The impacts of identity veriﬁcation and disclosure of social cues
on ﬂaming in online user comments
, K. Hazel Kwon
Pohang University of Science and Technology (POSTECH), Dept. of Industrial and Management Engineering, 77 Cheongam-ro, Nam-gu, Pohang, Gyeongbuk 790-784, South Korea
Arizona State University, School of Social and Behavioral Sciences, 4701 W. Thunderbird Rd. MC3051, Phoenix, AZ 85306, USA
Online public discussions
While a growing body of literature attests to the relationship between user identiﬁability and inﬂamma-
tory speech online, few studies have investigated the ways in which different anonymity control mech-
anisms affect the quality of online discussions. In this study, two mechanisms, a policy-driven and a
voluntary approach, are examined for their conditional and interaction effects on reducing ﬂaming in
user comments online. Based on a large-scale, real-world data on political news comments in South
Korea, the results suggest that whereas the policy-driven regulation does not reduce, and even increases,
ﬂaming, the voluntary approach signiﬁcantly decreases it, especially among the moderate commenters.
The ﬁndings are further speculated from an economic perspective by which transaction costs are per-
ceived differently contingent on the ways in which anonymous commenting is regulated.
Ó2015 Elsevier Ltd. All rights reserved.
The prevalent use of the Internet for political discussions has
brought forth the question of how online anonymity affects the
quality of public deliberation. On the one hand, scholars argue
for constructive roles of anonymity in increasing a sense of equal-
ity, reducing normative pressure on conformity, and encouraging
users to exert individual rights for free speech in a liberated man-
ner (Dahlgren, 2005; Papacharissi, 2004; Ruiz et al., 2011). On the
other hand, skeptics point out the detrimental effects of anonymity
on aggravating ﬂaming and trolling in online social interactions
(Kushin & Kitchener, 2009; Lampe, Zube, Lee, Park, & Johnston,
2014). Both positive and negative perspectives are rooted in the
same tenet that the degree of identiﬁcation can make a difference
in user behaviors.
As Herring et al. (2002) state, acrimonious comments, ‘‘while
clearly problematic, are nonetheless widespread and often toler-
ated, due in part to the pervasiveness on the Internet of civil liber-
tarian values that consider abusive speech a manifestation of
individual freedom of expression’’ (p. 372). The ease with which
the Internet anonymity may induce ﬂaming and trolling has
become a non-negligible issue especially along with growing con-
cerns about civility crisis in contemporary politics (Mutz & Reeves,
2005). Inﬂammatory political campaigns, for example, have
become pandemic in today’s news headlines and have been found
to delegitimize opposing point of views and political processes
(Brooks & Geer, 2007), and decrease citizens’ political trust (Mutz
& Reeves, 2005).
While some scholars may argue that online profanity may not
be as harmful as one might assume (Canter, 2013; Papacharissi,
2004), a recent survey indicates that the Internet and social media
were cited as one of the leading causes of incivility (Weber, 2013).
Coe, Kenski, and Rains (2014) also suggest the negative implication
of online profanity by showing that the more uncivil comments
(among which vulgarity and name-calling were subset) online
users are exposed to, the greater they become reluctant to engage
in discursive interactions. A study on Youtube ﬂaming similarly
suggests that the majority of users perceive ﬂaming ‘‘annoying’’
and ‘‘[deviating from] an honest way of expressing disagreement’’
(Moor, Heuvelman, & Verleur, 2010, p. 1542). Inasmuch as negative
responses elicited by ﬂaming are commonplace, it may be a valid
claim that ﬂaming in online comments could pose a danger to
the conditions for deliberative citizen-to-citizen discussions.
The relationship between ﬂaming and online anonymity has
drawn even more scholarly attention along with anonymity regu-
latory gestures recently put into high gear. Such gestures range
from promoting voluntary self-disclosures, to crowdsourcing mod-
eration, and to mandating identity veriﬁcation processes. In partic-
ular, the current study distinguishes two qualitatively different
mechanisms, one of which pertains to a policy-driven regulation
and the other to a voluntary disclosure of social cues, and explore
the extent to which each mechanism contributes to the reduction
0747-5632/Ó2015 Elsevier Ltd. All rights reserved.
Corresponding author. Tel.: +82 54 279 2375; fax: +82 54 279 2870.
E-mail addresses: email@example.com (D. Cho), firstname.lastname@example.org (K.H. Kwon).
Tel.: +1 602 543 5676; fax: +1 602 543 6004.
Computers in Human Behavior 51 (2015) 363–372
Contents lists available at ScienceDirect
Computers in Human Behavior
journal homepage: www.elsevier.com/locate/comphumbeh
of spiteful comments online. For empirical exploration, we lever-
age large, real-world data on news website comments in South
Korea where the Internet Identity Veriﬁcation Law was once
enforced nationwide (Cho & Kim, 2012). While we focus on
South Korea because its national law aptly exempliﬁes a
policy-driven, top-down approach in clear contrast to a voluntary
approach, we believe that this study should provide some general-
izable conclusions resonating with the intensiﬁed debates sur-
rounding anonymity regulations across different national and
The contribution of this study is twofold. First, while extant lit-
erature has examined the relationship between the Internet anon-
ymity and ﬂaming, few have differentiated the effects of dissimilar
anonymity control mechanisms. Thus, our study may provide addi-
tional theoretical as well as managerial insights for appropriating
different regulatory systems, when needed. Second, whereas the
related studies demonstrate analyses on a single site or company
(e.g. Coe et al., 2014;Lampe et al., 2014; Moor et al., 2010; Rowe,
2014), we leverage a large dataset beyond a particular platform:
The analyses cover online comments collected from 26 different
news websites, and the focus is on election campaign periods dur-
ing which exchanges of political opinions become increasingly cru-
cial for electorates’ decision making.
2. Literature reviews
2.1. Anonymity and ﬂaming in online comments
Although widespread use of online forums has facilitated plural
political debates, anonymity has been linked to the risk of depreci-
atory debate culture. Despite positive roles of the anonymous
Internet such as facilitating free speech without the fear of social
disapproval and even physical arrestment in sensitive political
contexts, anonymity has nevertheless provided leeway for spam-
ming, hate speech, deception and impersonation (Kling, Lee,
Teich, & Frankel, 1999). Moreover, ﬂaming and trolling have
recently become increasingly worrisome (Herring, Job-Sluder,
Scheckler, & Barab, 2002; Lampe et al., 2014).
In particular, ﬂaming refers to a message sender’s hostile emo-
tional expressions characterized by using insulting, profane, or
offensive languages, which may ‘‘inﬂict harm to a person or an
organization resulting from uninhibited behavior’’ (Alonzo &
Aiken, 2004, p. 205). On occasions, ﬂaming may be tolerable and
even liberating message senders by an emotional release.
However, every ﬂaming has a potential to negatively affect mes-
sage receivers: Although a sender might ﬂame for a harmless pur-
pose, the message can be perceived offensive depending on who
reads it (Moor et al., 2010). Empirical ﬁndings have shown that
ﬂaming can induce negatively emotional effects: Johnson, Cooper,
and Chin (2009), for example, suggest that ﬂaming downplay
computer-mediated negotiation processes by invoking angers.
Similarly, Coyne, Stockdale, Nelson, and Fraser (2011) ﬁnd that
media exposure to profanity, including the exposure to the
Internet content, induces more frequent uses of profane languages,
which subsequently increase physical and relational aggressive
behaviors among adolescents. Not just having individual psycho-
logical and emotional effects, ﬂaming may also play an unconstruc-
tive role in shaping online discussion culture: Verbal attacks can
become reciprocal, creating ‘‘ﬂaming norms’’ in an online commu-
nity (Lea, O’Shea, Fung, & Spears, 1992; Moor et al., 2010, p. 1542).
For example, Moor et al. (2010) contend that, despite the consen-
sus that ﬂaming is negative among users, inﬂammatory comments
have become increasingly normative in Youtube and such ﬂaming
norms cause some users’ withdrawal from content-generating
The underlying rationales behind ﬂaming are drawn from the
literature of anonymity effect on disinhibition. According to Scott
(1998), the concept of anonymity is a rich, multi-dimensional con-
cept, generally deﬁned as the condition to which a message
source’s true identity is absent or unacknowledged. The discus-
sions of anonymity can take a different direction depending on
whose perception is at the center of inquiry: self-anonymity refers
to a sender’s perceived anonymity to others, and other-anonymity
pertains to the anonymity a receiver experiences during an interac-
tion with unidentiﬁed source (Scott, 1998). This study focuses on
self-anonymity, also deﬁned as identiﬁability, because it directly
plays a role in determining the message sender’s own behaviors
(Spears & Lea, 1994). Accordingly, the concept ‘anonymity’ implies
self-anonymity hereupon and is used interchangeably with the
term identiﬁability. Self-anonymity effect on disinhibition is based
on the set of long-lasting theories on crowd behaviors and social
conformity (e.g. Festinger, Pepitone, & Newcomb, 1952; McPhail,
1991; Postmes & Spears, 1998; Zimbardo, 1969). The simplest
description of the rationale is: The less identiﬁable, the more disin-
hibited. Festinger et al. (1952) and Zimbardo (1969) introduce a
widely known concept called ‘‘de-individuation’’ to refer to the
extra- or anti-normative behaviors induced from the lack of per-
sonalized social cues.
Individuals become uninhibited because of the absence of
accountability. On a positive note, being free from the real-self
means less constraints, thus greater freedom and outspokenness
during discussion interactions (Jessup, Connolly, & Galegher,
1990; Zarsky, 2003). The downside, however, is that this tendency
can be translated into ill mannerism. Not surprisingly, negative
online activities such as defamation of character, threats, and slan-
derous comments have often been found problematic and become
a concern among law and policymakers (Cohen, 1995; Sunstein,
2014). A majority of empirical research on de-individuation has
indeed highlighted negative effects, particularly in relation to
aggressive, anti-normative behaviors (Christopherson, 2007). In
social media environment, different platforms are characterized
with different levels of affordances for identiﬁability and social
cues, which may have disproportionate effects on user disinhibi-
tion. For example, Halpern and Gibbs (2013) compare Facebook
and Youtube and ﬁnd that Youtube contains more impolite com-
ments due to its lower identiﬁability and networked information
access than Facebook.
The issue of ﬂaming has also been linked to the social identity
model of de-individuation effects (SIDE). Among various CMC the-
ories, this model is particularly suitable for analyzing the exchange
of commenting behaviors in the online discussion platforms where
‘‘one-off, non-interactive, pseudonymous messages are common’’
(Walther, DeAndrea, Kim, & Anthony, 2010, p. 473). Speciﬁcally,
the SIDE model proposes that reduced social cues in online settings
may actually facilitate in-group normative behaviors by allowing
communicators to treat one another equally as a member of the
online group (Sproull & Kielser, 1991) and trivializing individuated
unequal social status (Dubrovsky, Kielser, & Sethna, 1991).
De-personalization and increased conformity to a group norm,
which are the core dynamics of SIDE, occur when the process is
‘‘coupled with a salience of common (group) identity’’ (Lee, 2004,
p. 235; Postmes, Spears, Sakhel, & de Groot, 2001). While SIDE
may result in a stronger group cohesion and solidarity, the positive
consequence is not always the case. Indeed, some experimental
studies challenge the equalization hypothesis, claiming that
anonymous group communication online does not signiﬁcantly
increase equality among communicators of different status
(Coffey & Woolworth, 2004; Connolly, Jessup, & Valacich, 1990;
Hollingshead, 1996; Straus, 1997). Moreover, scholars even con-
tend that SIDE effect aggravates the tendency for group polariza-
tion (Sia, Tan, & Wei, 2002). For example, Lee’s (2007) study
364 D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372
shows that the increased group identity under anonymous condi-
tion induces subjects’ opinions to be more polarized. When ﬂaming
becomes a normative behavior within an online community, the
SIDE suggests that users motivation to conform to group norm
can result in more aggravated ﬂaming behaviors (Lea, O’Shea,
Fung, & Spears, 1992).
In sum, de-individuation thesis and SIDE model suggest that
anonymous conditions affect the ways in which individuals
behave. Although anonymity may help nurture a positive atmo-
sphere of open mindedness in certain situations, the same condi-
tion can likewise result in negative consequences in a different
context. Flaming is one aspect of online disinhibitions, which could
impose negative psychological effects as well as depreciate the cli-
mate of online public discussions.
2.2. Regulating anonymity: Policy-driven versus voluntary approaches
As anti-normative behaviors such as ﬂaming, trolling, and
cyber-threats have become commonplace online, growing attention
is being paid toward solutions that can increase user identiﬁability
and accountability. The scope of anonymity regulating attempts
vary in numerous ways: For example, many websites encourage
users to voluntarily link their social networking site (SNS) proﬁles
to the activities in the respective website (Rowe, 2014; Soni,
2013); Media companies often stipulate a policy that requires users
to provide real names during the registration, which often triggers
intricate issues on privacy and trust (Gross & Acquisti, 2005;
Stephen & Galak, 2012; Youmans & York, 2012); Furthermore, some
state governments and countries take more aggressive approaches
by mandating laws that make users’ real identities more easily trace-
able (Cho & Acquisti, 2013; Cho & Kim, 2012).
Such various attempts to control anonymity can be roughly cat-
egorized into two approaches: Policy-driven regulations, and pro-
motion of voluntary disclosure of social cues. (We do not
suggest, however, that the two approaches represent comprehen-
sive and mutually exclusive list of anonymity control mecha-
nisms.) Policy-driven regulations refer to enforcing formal rules
that compels users to be accountable for their online behaviors.
For example, companies like The Wall Street Journal and
require users to follow the real name policy, and theoret-
ically have the authority to block any users who do not comply with
the policy. On a state or national level, the government may either
require web service providers to collect users’ personal information,
as in the case of South Korea and China, or endorse law enforcement
authorities to gain access to the user data relatively easily, as exem-
pliﬁed by the recent legislature in Arizona (Newman, April 3, 2012).
The main purpose of policy-driven approaches is to increase trace-
ability of real identity: Even if users are allowed to use
pseudo-names online, users may not perceive self-anonymity at
least to surveillors of the website since each pseudo-name is trans-
latable into true identity based on personal information provided
(Kling et al., 1999). The premise behind implementing a
policy-driven approach as a solution for online disinhibition is that
the identity traceability would provoke an awareness of liability,
which would in turn prevent users from irresponsible actions online.
Alternatively, voluntary approach promotes a culture of
self-disclosure by taking advantage of recent social commenting
technologies (Wang, 2013). Conventionally, the use of social com-
menting system is offered as a more efﬁcient and convenient option
for users when they log into a respective website: Users can skip all
the sign-up steps, and immediately log into the website by simply
clicking a social plug-in that automatically connects the activities
on the website to an existing SNS account (Rowe, 2014). As a result,
identity-disclosing social cues displayed in the SNS proﬁles become
publicly visible on the website as well as the users’ comments,
which can likewise be shared by social contacts networked in the
SNS. Although such content ﬂuidity is likely to pose privacy con-
cerns and peer monitoring risks (Humphreys, 2011; Rainie &
Wellman, 2012), those who voluntarily opt in on the use of social
plug-ins tend to give more weight toward the immediate beneﬁts
(e.g. convenience, ease of use, and time saving) over the cost of los-
ing privacy (Smith, Dinev, & Xu, 2011). The underlying mechanism
behind the voluntary disclosure approach is to foster ‘‘public
self-awareness,’’ which refers to concerns about one’s appearances
and images constructed in social situations (Lee, 2007). The premise
behind social commenting approach as the remedy for online disin-
hibition is that public self-awareness evokes human desire for pos-
itive self-presentation. Motivated by social incentives associated
with positive display of self, users should practice mindfulness
when commenting in the networked environment.
In sum, various anonymity control mechanisms have been
widely implemented as an attempt to reduce anti-normative mes-
saging behaviors online. While policy-based and voluntary
approaches both aim for the common result – to reduce negative
behaviors – they each trigger subtly different sense of
self-anonymity: Policy-driven approaches activate a perception
that the self is not anonymous to the top-down monitoring body.
Such self-awareness may induce a sense of liability and obligation,
reminding users that they can be penalized for irresponsible
actions online. Alternatively, voluntary approaches accentuate
the identiﬁability to peer readers and evoke a social motives for
positive impression management, encouraging users to maintain
socially acceptable images online while avoiding irresponsible
actions that will compromise self-representation in the face of
increasing peer monitoring and subsequent social sanctions.
3. Research background and hypotheses
3.1. Research background
In order to reduce ﬂaming and trolling from discussion culture
on the Internet, the South Korean National Assembly enacted the
Identity Veriﬁcation Law in 2005, which was applicable to all polit-
ical comments made on major websites. This regulation required
service providers operating message boards on political websites
to permit users to post comments only after they had successfully
completed the identity veriﬁcation process (Park & Greenleaf,
2012). Although another moniker of this law was a ‘‘real-name’’
policy, the law did not explicate that users’ real names must be
publicized on a discussion forum. Therefore, the use of pseudonym
was allowed as long as the pseudonym is traceable to the ofﬂine
real identity. In December 2011, the regulatory agency ofﬁcially
announced that the law would be repealed in 2012 due to a possi-
ble harm and chilling effect on free speech (Park & Greenleaf,
2012). Moreover, the court’s ruling addressed that given the advent
of new means of communication including mobile messaging and
social networking sites (SNS), the identity veriﬁcation law would
only serve limited public interest across a narrow range of the
Internet space. Consequently, the Constitutional Court overturned
the law in August 2012.
History, as described above, indicates that General Election in
April 2012 was held during the transitional period preceding the
expiration of the law. Although the law was still in effect during
the two-week campaign period, some websites decided to no
longer require users to undergo the veriﬁcation process. Given
the random variation in the compliance level among websites dur-
ing this period, the division among compliant and non-compliant
websites was thought to qualify as the comparison groups. Set
Facebook recently shows the action to loosen the real name policy. See the report
in TechCrunch (January 30th, 2014) at http://www.techcrunch.com/2014/01/30/face-
D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372 365
aside the studies by Cho & Kim, 2012; Cho & Acquisti, 2013 that
demonstrated short-term effectiveness of the policy based on
two million comments across topical areas, little evidence exists
regarding the policy effect, in particular during a political election
campaign period. The investigation of political campaign period is
of particular interest because the top-down monitoring via identity
veriﬁcation system primarily aimed to deter Internet users from
political ﬂaming, with the assumption that such messaging should
harm the fairness in voters’ decision-making.
3.2. Research hypotheses
The following hypotheses are developed to evaluate the inﬂu-
ence of each anonymity control mechanism on user commenting
behaviors. In particular, we focus on the likelihood of ﬂaming as
a dependent variable. The contrast between the policy-based regu-
lation and the voluntary disclosure of social cues is at the center of
our examination. As previously mentioned, both approaches aim to
decrease anti-normative behaviors on the Internet. In particular,
policy-driven regulations may effectively control the level of anon-
ymity by increasing the traceability of real identity and subse-
quently increasing user accountability. Indeed, the intervention
to change disinhibited behaviors via increased accountability is
one of the main agendas for many existing Internet policies.
Therefore, our ﬁrst hypothesis is posited as follows:
H1. Formal policy regulations induce a sense of identiﬁability such
that commenters under the policy regulations are less likely to
write inﬂammatory comments than commenters who are not
under the policy regulations.
As for the effects of voluntary approaches, consider that social
motives are inﬂuenced by an individual’s belief on how he or she
is perceived by others. Comments that present the individual as
sociable, fair-minded, or caring may yield a positive image,
whereas comments that present the individual as unfair, aggres-
sive, or greedy may not only reduce the positive image but even
produce a negative image (Bénabou & Tirole, 2006). Thus, when
social cues are attached to the comments being posted, each indi-
vidual’s identiﬁable social image is at stake depending on the con-
tents of the comments. Consequently, settings where social images
are at stake may encourage commenters to avoid anti-normative
behaviors. This leads to our second hypothesis:
H2. Commenters who opt in on the voluntary disclosure of social
cues (i.e. social commenting) are less likely to write inﬂammatory
comments than commenters who did not choose a social com-
Lastly, hypotheses H1 and H2 are combined to assess how the
two suggested anonymity control mechanisms concurrently affect
commenting behaviors. The online commenting conditions in
which this study is conducted, described in further detail below,
allow us to test the effects of these two approaches together.
This hypothesis is formulated as follows:
H3. Commenters who use social commenting system under formal
policy regulations are even less likely to make inﬂammatory
comments than commenters who do not use social commenting
system under formal policy regulations.
4. Research design
4.1. Research setting
This study is based on the two-week General Election campaign
period in April 2012 in South Korea, which provides a natural,
real-world setting where the two aforementioned mechanisms
had co-existed to a varying degree across news websites. The
so-called Identity Veriﬁcation Law in South Korea had become
the target for debates around the world as an exemplar of
anti-anonymity governmental regulations.
In particular, concerns
were associated with its efﬁcacy on reducing trolling and ﬂaming
on the Internet. In addition to the identity veriﬁcation law in prac-
tice, many South Korean websites had adopted a social commenting
system, which allowed users to log into the website via their SNS
accounts. Data was acquired from the third-party company that con-
tinues to provide domestic news websites with a social commenting
system. This ﬁrm is the largest company of its kind in South Korea,
serving major media organizations since 2010. Thus, the study con-
text demonstrates an appropriate real-world setting with concur-
rence of the two anonymity control mechanisms.
Fig. 1 presents the 2 2 research design setting. The ﬁrst factor,
which measures whether or not commenters are members of the
news sites compliant with the law, represents a policy-driven
anonymity control mechanism. This factor categorizes sample
commenters into two groups: commenter from the
law-compliant websites and those from non-compliant websites.
When users wish to share comments on the law-compliant web-
sites, they must verify their real identity with a real name,
Residential Registration Number (a national identiﬁcation num-
ber), birthday, home address and phone number. In contrast, users
are not required to undergo such processes prior to posting mes-
sages on the non-compliant websites. Note that the veriﬁcation
process does not indicate that a commenter’s real name must be
publicly displayed. Rather, personal information is saved in the
website operators’ servers in order to enhance the identity trace-
ability and promote a sense of liability. The second factor repre-
sents the voluntary approach toward anonymity control. All of
the examined websites offer users an option of whether or not to
use a social commenting system. The amount of disclosed social
cues differs based on the user preference: Opting-in social com-
menting implies an agreement to disclose more social cues, in par-
ticular by connecting a SNS proﬁle with one’s commenting
behaviors. This process is consistent with the voluntary anonymity
control mechanism that encourages self-disclosure of social cues.
Therefore, the second factor categorizes sample commenters into
two groups: whether or not to opt-in a social commenting system.
A rich data set of online comments was collected in assistance
from the largest third-party social commenting provider in South
Korea. Our dataset contains a total of 13,219 political comments
written by 5753 commenters threaded under the articles in the
Politics and Election sections of 26 major news media sites. The
time window of data collection was during the two-week period
of the ofﬁcial election campaign (March 29–April 11). As shown
above, commenters and comments are distinguished by whether
the commenter’s personal identity was veriﬁed (Factor 1) and
whether an SNS account was used (Factor 2).
Although conceptualization of ﬂaming is often vague and
abstract, language uses have been at the center of its deﬁnition.
Therefore, in order to identify a dependent variable of the study,
comments that contain swearing or name-calling words were
deﬁned in this study as ﬂaming comments. In order to properly
The regulation pertains to Article 44-5, Section 1, Item 2 of the Act of Promotion
of Information and Communications Network Utilization and Information Protection
and Article 29 and Article 30, Section 1 of the Enforcement Ordinance of the same act.
In 2009, the law was strengthened to require websites having a daily viewership of
over 100,000 users to establish an identity-veriﬁcation mechanism. In the summer of
2012, the Constitutional Court overturned the law, calling it unconstitutional.
366 D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372
classify such comments, contents were reviewed based on a dic-
tionary of abusive languages. The dictionary consists of 651 com-
monly used strong swearwords and name-calling terms, which
are the variations of 319 words designated as abusive by a transna-
tional company Nielsen Korea, one of the largest audience and con-
sumer research ﬁrms in South Korea. The selected terms included
expletives and epithets (or commonly used pseudo-swearwords to
bypass automatic ﬁltering procedures) and other extreme
anti-normative terms frequently used in online communities in
South Korea. Two former journalists who worked at a major
South Korean news media organization reviewed the dictionary
to reassure the legitimacy of the dictionary. Exemplary ﬂaming
comments are translated into English for demonstration, with
‘‘These 6 year-old birdbrains, with no sense of the most basic
principle that public ofﬁcials should care about citizens’ lives,
keep avoiding their responsibility and are busy ﬁlling their
own stomachs. Do they deserve to be the congressmen of our
‘‘Min-Tong-Dang (one of the political parties in Korea) is a melt-
ing pot ﬁlled with bitches. How dare they nominate a bitch like
Mr. Kim? These bitches are commies in blood, and they don’t
care whether or not (a candidate is) an imposter, a liar, etc. as
long as he has the commie taste.’’
‘‘President Rho (one of the former presidents in Korea) assisted
your ‘beloved’ North Korea to arm with nukes. He should have
been impeached at least 100 times. You retards need to get a
sense back...or get hit by nuke!’’
Inﬂammatory comments were automatically classiﬁed with the
assistance of software. To verify the validity of our classiﬁcation
approach using the dictionary, 300 comments in the subset of
ﬂaming comments were randomly selected, and two human coders
independently and manually reviewed each comment to deter-
mine if the comment is properly classiﬁed. Both coders agreed that
over 97% of comments can legitimately be regarded as ﬂaming
comments by containing swearing terms, and a Cohen’s kappa
score of 0.82 suggested sufﬁcient inter-coder reliability.
In addition, comment- and commenter-speciﬁc variables
that may affect commenting behaviors were identiﬁed, and the
descriptive statistics were produced, which are presented in
5. Empirical analysis and results
As shown in Table 2, the majority of commenters used an SNS
account (N= 3908, 67.93%) and the majority of comments were
written in the news sites that were non-compliant with the identity
veriﬁcation law (N= 9666, 73.12%). This may be due to the fact that
social commenting system was widely adopted due to its conve-
nience, and that more news sites in our sample (21 out of 26 sites)
were non-compliant sites. Despite unequal sample size, the website
characteristics (e.g. organizational scales and political predisposi-
tions) varied randomly, eliminating the risk for systematic biases.
Using the 2 2 taxonomy described in Table 2, the ratios of
inﬂammatory comments are calculated: Given that an SNS account
is used, there is a higher likelihood of including ﬂaming when com-
ments are subject to the identity veriﬁcation law. Given that a
website is compliant with the law, however, there is a lower like-
lihood of including inﬂammatory comments when an SNS account
is used (Fig. 2).
For formal hypotheses tests, we utilize the random effects panel
Probit models (Poirier, 1980) in order to estimate the effects of
treatments on the probability of ﬂaming while acknowledging that
the comments written by a single commenter may not be indepen-
dent of one another. In other words, by establishing the unit of
analysis as a commenter, where each comment constitutes a sepa-
rate yet correlated data point, we control for the unobservable fac-
tors speciﬁc to each commenter. This approach assumes constant
correlation across all comments written by a single commenter
and accounts for that correlation in the variance–covariance matrix
of the coefﬁcients. While the estimated coefﬁcients do not repre-
sent marginal effects of independent variables on the likelihood
of inﬂammatory comments, the directionality of the effect esti-
mated by the coefﬁcient is important to note. Similarly, the magni-
tude of the marginal effect is not the primary focus of this study,
although the true effect would be proportional to the magnitude
of the presented coefﬁcient. To highlight the effects of veriﬁcation
law and social commenting (Hypotheses H1–H3), the following
regression frame is used:
where iindexes the commenter and jindexes the comment. The
dependent variable, Flaming, is equal to one if the comment
includes designated swearing terms, and zero otherwise.
Similarly, Veri is a dummy variable that indicates whether or not
Fig. 1. Summary of research setting. Note: The identity-veriﬁcation law is applied to newspapers in the Group 1 through which it is required to verify a real identity by
submitting a national identiﬁcation number. The law is not applied to newspapers in the Group 2. Commenters then choose an account between SNS and non-SNS accounts.
SNS accounts include Facebook, Twitter, and other local SNS services. A non-SNS account includes a news site account. If a commenter signs in with any of SNS accounts, a
user’s current SNS proﬁle picture that often contains a person’s face image or other information that may be connected to a user’s real identity. If others are interested in the
commenter’s proﬁle, they can visit the commenter’s personal SNS webpage by simply clicking the displayed image. If a user chooses a non-SNS account, the newspaper logo is
appeared instead of a user’s SNS proﬁle picture. Followed by the image, a user name is displayed, which can be either a real name or a nickname.
D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372 367
a comment is subject to identity veriﬁcation, and SNS denotes
whether or not a comment is produced via SNS account.
AllComments and AllLengths are discrete variables that control for
commenter-speciﬁc characteristics and represent the number of
all comments and the total comment length by a single commenter
during the study period, respectively.
The unobserved variables,
denoted by the error terms, are assumed to be normally distributed.
The variables of interest are Veri (or SNS), respective to hypothe-
ses H1 (or hypothesis H2), and Veri SNS, an interaction term
between Veri and SNS. A negative coefﬁcient on Veri (or SNS) would
suggest that commenters are less likely to include ﬂaming under
the identity veriﬁcation law (or under the use of an SNS account).
A coefﬁcient on Veri SNS corresponds to hypothesis H3 and rep-
resents the compounding effect of the law and an SNS account.
The results of this study are displayed in Table 3. First, the pos-
itive and statistically signiﬁcant coefﬁcient for Veri in Column (1)
suggests that the implementation of the law increases the likeli-
hood of ﬂaming. This result does not support the hypothesis H1
and is actually opposite of the policy’s original objective.
Conversely, the negative and statistically signiﬁcant coefﬁcient
for SNS in Column (2) indicates that the use of an SNS account is
associated with decrease in ﬂaming, which supports hypothesis
H2. The results from Column (3) include variables Veri and SNS
as well as the interaction term Veri SNS, and suggest that the esti-
mated coefﬁcient of Veri remains signiﬁcant and positive while the
coefﬁcient on the interaction term is identiﬁed as negative and sig-
niﬁcant. This ﬁnding indicates that although the identity veriﬁca-
tion law is associated with elevated probability of ﬂaming, the
use of SNS accounts could counterbalance such likelihood and alle-
viate inﬂammatory commenting (Hypothesis H3 is supported).
Lastly, the results from the control variables indicate that (1) there
is a low likelihood of ﬂaming as a commenter writes more com-
ments, and (2) the length of comments is positively associated
with the likelihood of ﬂaming.
The analyses presented in the previous section assume random
distribution of commenters across news sites as well as random
distribution of SNS account users versus non-users. This assump-
tion leaves room for a potential bias in which the composition of
a particular group of commenters may be systematically different
from that of other groups—so-called the selection bias. In
particular, the signiﬁcant results of commenter characteristics
(AllComments and AllLengths) may invoke suspicions regarding
any systematic bias induced by the characteristics of commenters.
To address this concern, additional modeling is conducted based
on the commenter types. Speciﬁcally, the commenters are
categorized into three groups – heavy, moderate, and light
commenters – in order to account for commenter characteristics,
which are then used in propensity score matching (PSM) (Dehejia
& Wahba, 2002; Rosenbaum & Rubin, 1983). The PSM method
provides a randomized, experiment-like setting that excludes the
impact of unobserved heterogeneity by matching observational
commenter-speciﬁc characteristics between the two groups of
observations. We formally present the model as follows:
EðFlaming ¼1jTreatment ¼1;XÞ>EðFlaming ¼1jTreatment ¼0;XÞ
where the treatment groups are deﬁned as commenters subject to
the identity-veriﬁcation law or those who use an SNS account. X
is a vector covariate including AllComments and AllLengths. The vari-
able, SNS (or Veri), is included as one of covariates when the treat-
ment is the Veri (or the SNS).
The average treatment effects (ATE) using the PSM method are
illustrated in Tables 4 and 5.
First, Table 4 takes a close-up look at
ﬂaming occurrences by comparing the composition of commenters
and comments between identity veriﬁed and non-veriﬁed groups.
As seen in the table, the heavy user ratio of commenter and comments
in the identity veriﬁed group is much larger than that in the
non-veriﬁed group. Furthermore, the proportion of ﬂaming comments
in the identity veriﬁed group is larger than that in the non-veriﬁed
group. The ATE from the PSM test show that all group-speciﬁc coefﬁ-
cients are positive and statistically signiﬁcant, suggesting that the
implementation of the identity veriﬁcation law is indeed associated
with high probability of inﬂammatory language uses.
Table 5 compares between SNS account users and non-SNS
account users. While the number of commenters is highly skewed
to the group of light commenters, no signiﬁcant difference is
observed in terms of the proportion of ﬂaming comments among
Variable Description Mean Std. dev. Min Max
languages or not
0.1353 0.3421 0 1
Veri Subject to the
0.2687 0.4433 0 1
SNS Use of SNS
account or not
0.5954 0.4908 0 1
AllComments The number of
15.7737 36.0806 1 233
AllLengths All lengths of
1226.9110 2383.8260 1 14,450
The number of comments and commenters.
Factor 1 Sum
to the law
to the law
Factor 2 SNS 1568 (570) 6303 (3338) 7871 (3908)
Non-SNS 1985 (348) 3363 (1497) 5348 (1845)
Sum 3553 (918) 9666 (4835) 13,219 (5753)
Law Yes No Yes No
SNS Yes Yes No No
Mean = 0.1353
Fig. 2. The ratio of comments including ﬂaming.
We assume that commenters are randomly distributed across news sites in our
sample, and news sites are randomly subject to the identity-veriﬁcation law. These
commenter-speciﬁc variables are included for the descriptive purpose to see how the
type of commenters is related to offensive behaviors.
We use the PSMATCH, propensity score matching module, in Stata 13 to match
samples using the nearest 10 neighbors.
368 D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372
light commenters. It is possible that the light commenters may write
only one or a few comments extemporaneously when they want to
express their opinions. Noticeable differences, however, are
observed in terms of the proportion of ﬂaming comments within
moderate and heavy commenter groups, suggesting that moderate
and heavy commenters may care about their social images repre-
sented by their commenting behaviors more than light commenters.
Despite the ostensible difference of the proportion of ﬂaming within
the heavy commenter group, however, the difference is not so statis-
tically signiﬁcant after we control possible covariates (p-value:
0.053). This is due to the fact that some heavy commenters using a
non-SNS account wrote the large number of ﬂaming comments,
whereas those using an SNS account wrote the very small number
of ﬂaming comments – for example, a heavy commenter using
non-SNS account wrote 32 ﬂaming comments out of 48 comments
(66.7%), while a heavy commenter using an SNS account wrote no
ﬂaming comments out of 119 comments (0.0%). PSM results control
these possible outliers, and focus on the likelihood of writing at least
one ﬂaming comment at the commenter-level. As a result, all
group-speciﬁc coefﬁcients are negative but not all coefﬁcients are
statistically signiﬁcant, indicating that the use of an SNS account
may decrease the probability of inﬂammatory comments especially
for the moderate-level commenters.
6. Discussion and conclusions
The issue of anonymity has been at the center of scholarly dis-
courses on ways to promote respectful climate for online public
The effects of identity veriﬁcation law and social commenting on ﬂaming.
Column (1) (2) (3)
Coefﬁcient Std. error Coefﬁcient Std. error Coefﬁcient Std. error
(0.0875) 0.0337 (0.1009)
Veri SNS 0.5405
All Comments 0.0176
All Lengths 0.0001
0.000 0.000 0.000
# of commenters 5753 5753 5753
Number of comments 13,219 13,219 13,219
Robust standard errors are in parentheses.
The proportion of commenter, comment, and ﬂaming between identity veriﬁed and non-veriﬁed group and propensity score matching test results (DV = Flaming).
Subject to the law Not subject to the law ATE
Proportion of ﬂaming
Proportion of ﬂaming
(1) Light commenter 705 (76.8%) 1047
15.4 4364 (90.3%) 5637
142 (15.5%) 791 (22.2%) 20.4 377 (7.8%) 1990
71 (7.7%) 1715
18.5 94 (1.9%) 2039
(4) Sum 918 (100.0%) 3553
Note: Light commenters are those who left 1–3 comments, moderate who left 4–9, and heavy commenter who left more than or equal to 10; ATE = Average Treatment Effect.
Robust standard errors are in parentheses under ATE.
The proportion of commenter, comment, and ﬂaming between SNS and Non-SNS account users and propensity score matching test results (DV = Flaming).
SNS account users Non-SNS account users ATE
Proportion of ﬂaming
Proportion of ﬂaming
(1) Light commenter 3521 (90.1%) 4617
13.0 1548 (83.9%) 2067
316 (8.0%) 1685
13.5 203 (11.0%) 1096
71 (1.9%) 1569
5.6 94 (5.1%) 2185
(4) Sum 3908
Note: Light commenters are those who left 1–3 comments, moderate who left 4–9, and heavy commenter who left more than or equal to 10; ATE = Average Treatment Effect.
Robust standard errors are in parentheses under ATE.
D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372 369
discussions. In line with recent regulatory efforts to moderate
anonymous commenting behaviors, this study explores the effects
of different anonymity control mechanisms – policy-driven regula-
tions and the promotion of voluntary self-disclosure – on inhibit-
ing inﬂammatory comments. The policy-driven approach
accentuates a sense of identiﬁability by the entities who are in
charge of penalizing anti-normative behaviors, while the voluntary
approach appeals to the identiﬁability by peer users, which per-
tains to the social incentives of positive self-presentation. To com-
pare these two mechanisms, this study leverages a large,
real-world dataset of newspaper website comments during an
election campaign period in South Korea. The particular societal
context is chosen due to the natural real-world setting created
through the coexistence of the nationwide identity veriﬁcation
law (a policy-driven regulatory approach) and the widespread
use of social commenting system (a voluntary approach).
Verifying identiﬁcation credentials do not indicate that com-
menters’ real name is displayed with their comments. Instead, as
far as a user’s identity is veriﬁed through the system of the website
operator, it allowed users to maintain pseudonyms or screen
names when they write their comment. Using an SNS account,
however, indicates that a commenter’s social cues are likely to be
disclosed, because others can visit the commenter’s personal SNS
webpage. This study reveals that the enforcement of identity veri-
ﬁcation law is noticeably associated with an increased probability
of inﬂammatory commenting, which is counterintuitive against
the proposed policy goal. On the contrary, the use of an SNS
account is negatively correlated with ﬂaming. Our results suggest
that commenters are more likely to be affected by the disclosure
of their SNS proﬁle with a comment than are they subject to the
law. In other words, while both a policy-driven regulation and a
voluntary disclosure of social cues are mechanisms to enhance
the degree of identiﬁability, the public visibility of personal proﬁle
through using an SNS account tends to be related to reducing ﬂam-
ing in user comments online.
Additional PSM tests on commenter groups of different levels of
activities (heavy, moderate, and light commenters) validate that
the effects of the two mechanisms uniquely account for ﬂaming
across commenters with varying commenting behaviors. In partic-
ular, social commenting effect was the most prominent among the
moderate commenters. This may be due to the fact that more or
less frequent commenters may perceive their comments to affect
their social image and be aware of language choices when they
use an SNS account. We also observed a salient difference of the
proportion of ﬂaming comments between using an SNS account
and a non-SNS account for heavy commenters, but the difference
seems to result from a few outliers within the group of heavy
6.1. Theoretical implications
Considering that the identity veriﬁcation law in South Korea
had not required public display of real names, the ostensible anon-
ymity, due to the use of pseudonyms, and the salient common
political identity among highly opinionated likeminded users
could have created a combined effect on inducing anti-normative
commenting against out-group entities. This rationale is consistent
with the SIDE model yet resonates with negative consequences. If
this dynamic is truly occurring, the identity veriﬁcation policy, and
possibly any other top-down regulations that increase the transac-
tion cost for political expressions, could prompt unfavorable condi-
tions of public deliberations such as aggravation of political
polarization and removal of moderate voices from the public opin-
ion landscape. While this study could not investigate how the ver-
iﬁcation process – or other policy implementation that increase the
transaction cost of discussion engagement – is related with the
salience of group identity, future research is recommended to
delve into the role of group political identity in polarization and
Our results of reduced ﬂaming comments by using an SNS
account are in line with ﬁndings in a majority of empirical research
on deindividuation effect, indicating that the more identiﬁable, the
less disinhibited. The voluntary disclosure of social cues is also
related to social penetration theory, suggesting that increasing
degree of self-disclosure is an outcome of natural evolution of
interpersonal relationships (Altman & Taylor, 1973). As far as the
minimum level of privacy is placed, privacy provides a foundation
for self-disclosure, which allows users to build social capital and
self-esteem and to achieve the desire for interaction, socialization,
and recognition (Acquisti, Brandimarte, & Lowenstein, 2015).
Commenters using an SNS account may believe that these motives
are more essential than the protection of privacy, and they are
likely to behave prosocially on the Internet in order not to produce
a negative social image (Bénabou & Tirole, 2006).
6.2. Practical implications
Our results also have several policy and managerial implica-
tions. Most importantly, our ﬁndings suggest that the effectiveness
of top-down interventions that increase the traceability of identity
may be limited or even counterproductive on shaping political dis-
cussion culture. The results regarding the Identity Veriﬁcation Law
suggest that some potential commenters may withdraw them-
selves from writing comments due to the inconvenience and risks
associated with the veriﬁcation process, whereas others who
undergo the veriﬁcation process may do so because they perceive
the beneﬁts gained from becoming an online community member
and expressing opinions to be higher than the costs incurred from
the veriﬁcation process. In such scenario, it is possible that the
comments posted on a particular news site could over-represent
the highly opinionated individuals who share political ideology
promoted by the reader community of the news website.
This conjecture could be true especially in the South Korean
context where most major newspapers reveal political preferences
to a certain extent. Highly opinionated comments among like-
minded users could polarize the political discussion climate,
resulting in greater verbal aggressions against opposite opinions
or individuals. The heightened transaction cost induced by the
identity veriﬁcation process could encourage such a pattern,
resulting ﬂaming towards non-likeminded others to be taken as a
normative behavior within the likeminded community. The com-
parison of ﬂaming proportions between identity veriﬁed and
non-veriﬁed groups, as seen in Table 4 above, supports our infer-
ence in that the identity veriﬁed users are more likely to repeatedly
write inﬂammatory comments and become aggressive against the
While the Korea’s Identity Veriﬁcation Law may be regarded as
one of a few special cases with heavy hand in regulating the
Internet space by the government, preventing malicious use of
the Internet has been of key issue in many countries. For example,
the use of surveillance-oriented information technologies, of which
the Internet monitoring is a central part, for national security is a
growing phenomenon over the world. Accordingly, the enhanced
level of identity traceability on the Internet has become a common
practice across many governments. However, at the same time, the
civil rights advocates worry that legal actions to eradicate the
Internet privacy for the sake of security may affect online freedom
of speech. Our results imply that forceful governmental interven-
tion by implementing the veriﬁcation process may not be a legiti-
When it comes to the private sector, Google had an intention to
force users to use their real name as they launched a new SNS,
370 D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372
Google+, in 2012, and it reversed the policy in 2014 after the sub-
stantial criticism. Facebook still holds the real name policy, and
advocacy groups, such as the Electronic Frontier Foundation, keep
challenging the policy, because the Facebook’s policy led to non-
trivial ongoing disputes, such as recent controversies against
Native Americans and LGBTQ communities. These can be another
evidences that the real name policy may provoke substantial prob-
lems, which is in accordance with our ﬁnding. Given that the issue
of real-identity veriﬁcation on the Internet is still controversial in
various aspects, our study may provide practical implications to
those countries and online service providers involved in the mat-
ters, suggesting that implementing the identity veriﬁcation process
as a remedy for the potential threat could result in unexpected pol-
icy outcomes. On another note, the positive association between
the implementation of the law and ﬂaming could also mean that
the citizens have become desensitized to the regulations and thus
no longer care about liability. This interpretation is consistent with
the ﬁndings of short-term effects of the Identity Veriﬁcation Law
by the previous studies (Cho & Kim, 2012). Our results point out
a fundamental top-down policy problem inherent in the Internet
laws in general: As the use of social media becomes widespread
at an accelerating speed, bureaucratic control of the far-reaching
Internet space becomes extremely difﬁcult and nearly impossible.
Meanwhile, our ﬁndings provide strong evidence for the possi-
bility of effective implementation of voluntary means, in particular
a social commenting system. The spillover of social cues seems to
motivate users to express themselves in a responsible manner.
Furthermore, users may choose to utilize the social commenting
system because it lowers the transaction cost and increases conve-
nience through bypass of the sign-up process. This is contrary to
the identity veriﬁcation process, and may possibly facilitate spon-
taneous participation in political discourses without the burden of
establishing oneself as a member of the community. The current
ﬁndings show the potential beneﬁts of implementing voluntary
approach in reducing anti-normative messaging.
However, it remains to be answered by future research whether
the voluntary approach only reduces negative commenting behav-
iors or suppresses all types of discussion activities as a whole.
There are anecdotal evidences suggesting that the disclosure of
social cues does not necessarily decrease the volume of comments.
For example, Hufﬁngton Post, which implements a social com-
menting system, has experienced steadfast increase of comments
from 3 million in 2011 to 9 million in 2013 (Soni, 2013).
However, descriptive examples do not validate a signiﬁcant associ-
ation between the voluntary approach and the general liveliness of
the discussion culture. More systematic investigation is recom-
mended for delving into broad, long-term implications of the
self-disclosure culture on promoting public discussions.
Furthermore, the voluntary mechanism potentially provokes other
social issues such as privacy. Although discussion about such
related issues is beyond the scope of the current study, these issues
must be addressed by future research.
That said, we note that the study has several limitations. First
and most importantly, we operationalized ﬂaming somewhat nar-
rowly, by classifying a comment as ﬂaming if it includes an abusive
language. Accordingly, a possibility to include false negative cases
exists: The classiﬁcation could neglect comments that are subtly
abusive without the use of profane languages. While we used a
computer-assisted classiﬁcation approach due to the data size,
future research might be interested in more qualitative, granular
approach to explore the ways in which ﬂaming is expressed in user
commenting. Second, although we have attempted to address dif-
ferences across news sites with and without the enforcement of
the Identity Veriﬁcation law as well as differences among com-
menters using an SNS account or a non-SNS account, we cannot
completely rule out the possibility of selection bias, as is the case
with any observational study. To strengthen the external validity,
further investigations using an addition data set – for example,
data from the non-election period after the expiration of the law,
or data containing comments from news sites where the voluntary
mechanism is not available – may provide additional insights. In
addition, possible concerns exist on whether the news sites
selected in the study could be biased. Nonetheless, we believe that
our sample is an appropriate source from which to assess com-
menting behaviors of the population because our sample shows a
sufﬁcient variation in terms of types of readers, focus (e.g., busi-
ness newspapers and local newspapers), and political orientations.
Despite some limitations, this study contributes to understand
different anonymity moderation mechanisms in online news com-
munities and how these mechanisms affect to reduce ﬂaming in
citizen-to-citizen public discussion. A policy-driven process could
result in side effects by increasing transaction costs, while social
commenting approach could facilitate discussion engagements in
a respectful manner by decreasing transaction costs. The explo-
ration of a large-scale real-world data has a unique contribution
to understand the relationship between anonymity and ﬂaming.
Although limited factors were investigated in this study, the study
does not dismiss other effects not included in the analyses. More
extensive investigations on the content and valence of messages
may help reveal new insights beyond the scope of this study.
Acquisti, A., Brandimarte, L., & Loewenstein, G. (2015). Privacy and human behavior
in the age of information. Science, 347(6221), 509–514.
Alonzo, M., & Aiken, M. (2004). Flaming in electronic communication. Decision
Support Systems, 36(3), 205–213.
Altman, I., & Taylor, D. (1973). Social penetration: The development of interpersonal
relationships. New York: Holt, Rinehart & Winston.
Bénabou, R., & Tirole, J. (2006). Incentives and prosocial behavior. American
Economic Review, 96(5), 1652–1678.
Brooks, D. J., & Geer, J. G. (2007). Beyond negativity: The effects of incivility on the
electorate. American Journal of Political Science, 51, 1–16.
Canter, L. (2013). The misconception of online comment threads. Journalism
Practice, 7(5), 604–619.
Cho, D., & Acquisti, A. (2013). The More social cues, the less trolling? An empirical
study of online commenting behavior. Working paper.
Cho, D., & Kim, S. (2012). Empirical analysis of online anonymity and user
behaviors: the impact of real name policy. In Proceeding of the 2012 45th Hawaii
international conference on system sciences (HICSS) (pp. 3041–3050).
Christopherson, K. M. (2007). The positive and negative implications of anonymity
in Internet social interactions: ‘On the Internet, nobody knows you’re a dog’.
Computers in Human Behavior, 38(6), 3038–3056.
Coe, K., Kenski, K., & Rains, S. A. (2014). Online and uncivil? Patterns and
determinants of incivility in newspaper website comments. Journal of
Communication, 64(4), 658–679.
Coffey, B., & Woolworth, S. (2004). ‘‘Destroy the scum, and then neuter their
families:’’ the web forum as a vehicle for community discourse? The Social
Science Journal, 41(1), 1–14.
Cohen, J. E. (1995). Right to read anonymously: A closer look at copyright
management in cyberspace. Connell Law Review, 28, 981–1040.
Connolly, T., Jessup, L. M., & Valacich, J. S. (1990). Effects of anonymity and
evaluative tone on idea generation in computer-mediated groups. Management
Science, 36(6), 689–703.
Coyne, S. M., Stockdale, L. A., Nelson, D. A., & Fraser, A. (2011). Profanity in media
associated with attitudes and behavior regarding profanity use and aggression.
Pediatrics, 128(5), 867–872.
Dahlgren, P. (2005). The Internet, public spheres, and political communication:
Dispersion and deliberation. Political Communication, 22(2), 147–162.
Dehejia, R. H., & Wahba, S. (2002). Propensity score matching methods for
nonexperimental causal studies. The Review of Economics and Statistics, 84(1),
Dubrovsky, V., Kiesler, S., & Sethna, B. (1991). The equalization phenomenon: Status
effects in computer-mediated and face-to-face decision-making groups.
Human–Computer Interaction, 6(2), 19–146.
Festinger, L., Pepitone, A., & Newcomb, T. (1952). Some consequences of de-
individuation in a group. Journal of Abnormal and Social Psychology, 47(2s),
Gross, R., & Acquisti, A. (2005). Information revelation and privacy in online social
networks. In Presented at the Workshop on Privacy in the Electronic Society, New
York, NY (pp. 71–80).
Halpern, D., & Gibbs, J. (2013). Social media as a catalyst for online deliberation?
Exploring the affordances of Facebook and YouTube for political expression.
Computers in Human Behavior, 29(3), 1159–1168.
D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372 371
Herring, S., Job-Sluder, K., Scheckler, R., & Barab, S. (2002). Searching for safety
online: Managing ‘‘trolling’’ in a feminist forum. The Information Society, 18(5),
Hollingshead, A. B. (1996). Information suppression and status persistence in group
decision making: The effects of communication media. Human Communication
Research, 23(2), 193–219.
Humphreys, L. (2011). Who’s watching whom? A study of interactive technology
and surveillance. Journal of Communication, 61, 575–595.
Jessup, L. M., Connolly, T., & Galegher, J. (1990). The effects of anonymity on GDSS
group process with an idea-generating task. MIS Quarterly, 14(3), 313–321.
Johnson, N. A., Cooper, R. B., & Chin, W. W. (2009). Anger and ﬂaming in computer-
mediated negotiation among strangers. Decision Support Systems, 46(3),
Kling, R., Lee, Y. C., Teich, A., & Frankel, M. S. (1999). Assessing anonymous
communication on the internet: Policy deliberations. The Information Society,
Kushin, M. J., & Kitchener, K. (2009). Getting political on social network sites:
Exploring online political discourse on Facebook. First Monday, 14(11). http://
Lampe, C., Zube, P., Lee, J., Park, C. H., & Johnston, E. (2014). Crowdsourcing civility:
A natural experiment examining the effects of distributed moderation in online
forums. Government Information Quarterly, 31(2), 317–326.
Lea, M., O’Shea, T., Fung, P., & Spears, R. (1992). ‘Flaming’ in computer-mediated
communication: Observations, explanations, implications. In M. Lea (Ed.),
Contexts of computer-mediated communication (pp. 89–112). London:
Lee, E. J. (2004). Effects of visual representation on social inﬂuence in computer-
mediated communication. Human Communication Research, 30(2), 234–259.
Lee, E. J. (2007). Deindividuation effects on group polarization in computer-
mediated communication: The role of group identiﬁcation, public-self-
awareness, and perceived argument quality. Journal of Communication, 57(2),
McPhail, C. (1991). The myth of madding crowd. New York, NY: Walter de Gruyter.
Moor, P. J., Heuvelman, A., & Verleur, R. (2010). Flaming on Youtube. Computers in
Human Behavior, 26(6), 1536–1546.
Mutz, D. C., & Reeves, B. (2005). The new videomalaise: Effects of televised incivility
on political trust. American Political Science Review, 99, 1–16.
Newman, J. (April 3, 2012). Arizona looks to outlaw Internet trolling. Time. <http://
Papacharissi, Z. (2004). Democracy online: Civility, politeness, and the democratic
potential of online political discussion groups. New Media & Society, 6(2),
Park, W., & Greenleaf, G. (2012). Korea rolls back ‘real name’ and ID number
surveillance. University of New South Wales Faculty of Law Research Working
Poirier, D. J. (1980). Partial observability in bivariate probit models. Journal of
Econometrics, 12, 209–217.
Postmes, T., & Spears, R. (1998). Deindividuation and anti-normative behavior: A
meta-analysis. Psychological Bulletin, 123, 238–259.
Postmes, T., Spears, R., Sakhel, K., & de Groot, D. (2001). Social inﬂuence in
computer-mediated communication: The effects of anonymity on group
behavior. Personality and Social Psychology Bulletin, 27, 1243–1254.
Rainie, L., & Wellman, B. (2012). Networked: The new social operating system.
Cambridge, MA: MIT Press.
Rosenbaum, P., & Rubin, D. (1983). The central role of the propensity score in
observational studies for causal effects. Biometrika, 70(1), 41–55.
Rowe, I. (2014). Civility 2.0: A comparative analysis of incivility in online political
discussion. Information, Communication & Society.http://dx.doi.org/10.1080/
Ruiz, C., Domingo, D., Micó, J. L., Díaz-Noci, J., Meso, K., & Masip, P. (2011). Public
sphere 2.0? The democratic qualities of citizen debates in online newspapers.
The International Journal of Press/Politics, 16(4), 463–487.
Scott, C. (1998). To reveal or not to reveal: A theoretical model for anonymous
communication. Communication Theory, 8(4), 381–407.
Sia, C.-L., Tan, B. C. Y., & Wei, K.-K. (2002). Group polarization and computer-
mediated communication: Effects of communication cues, social presence, and
anonymity. Information Systems Research, 13(1), 70–90.
Smith, H. J., Dinev, T., & Xu, H. (2011). Information privacy research: An
interdisciplinary review. MIS Quarterly, 35(4), 989–1016.
Soni, J. (August 26, 2013). The reason HuffPost is ending anonymous accounts.
Spears, R., & Lea, M. (1994). Panacea or panopticon? The hidden power in computer-
mediated communication. Communication Research, 21(4), 427–459.
Sproull, L., & Kielser, S. (1991). Connections: New ways of working in the networked
organization. Cambridge, MA: MIT Press.
Stephen, A. T., & Galak, J. (2012). The effects of traditional and social earned media
on sales: A study of a microlending marketplace. Journal of Marketing Research,
Straus, S. (1997). Technology, group process, and group outcomes: Testing the
connections in computer-mediated and face-to-face groups. Human–Computer
Interaction, 12(3), 227–266.
Sunstein, C. R. (2014). On rumors: How falsehoods spread, why we believe them, and
what can be done. Princeton, NJ: Princeton University Press.
Walther, J. B., DeAndrea, D., Kim, J., & Anthony, J. C. (2010). The inﬂuence of online
comments on perceptions of antimarijuana public service announcements on
YouTube. Human Communication Research, 36(4), 469–492.
Wang, S. S. (2013). ‘‘I Share, Therefore I Am’’: Personality traits, life satisfaction, and
Facebook Check-Ins. Cyberpsychology, Behavior, and Social Networking, 16(12),
Weber, & Shandwick (2013). Civility in America 2013. <http://www.
Summary.pdf> (Retrieved December 2014).
Youmans, W. L., & York, J. C. (2012). Social media and the activist toolkit: User
agreements, corporate interests, and the information infrastructure of modern
social movements. Journal of Communication, 62(2), 315–329.
Zarsky, T. Z. (2003). Thinking outside the box: Considering transparency,
anonymity, and pseudonymity as overall solutions to the problems in
information privacy in the Internet society. University of Miami Law Review,
Zimbardo, P. G. (1969). The human choice: Individuation, reason, and order versus
deindividuation, impulse, and chaos. Nebraska symposium on motivation (Vol.
17, pp. 237–307). University of Nebraska Press.
372 D. Cho, K.H. Kwon / Computers in Human Behavior 51 (2015) 363–372