Available via license: CC BY-NC 4.0
Content may be subject to copyright.
https://doi.org/10.1177/20563051211064906
Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-
NonCommercial 4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction
and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages
(https://us.sagepub.com/en-us/nam/open-access-at-sage).
Social Media + Society
October-December 2021: 1 –17
© The Author(s) 2021
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/20563051211064906
journals.sagepub.com/home/sms
Article
Introduction
Facebook Groups can contain a variety of topics, from health
and hobbies to religion and politics, and they can also pro-
mote a specific event or act as a marketplace. Participation is
often used as a synonym for civic engagement and thus refers
to a broader area of topics including (but not restricted to)
politics or the public sphere. It also includes what scholars
have termed cultural engagement (Lutz et al., 2014). Such
participation is a crucial force in shaping contemporary soci-
eties, and gender equality in engagement secures human
rights and social justice (Krook & True, 2012).
Studies on general digital communication effects reveal a
systemic gender inequality, with women being less civically
engaged than men both off- and online (Fortin-Rittberger,
2016; Fraile, 2014). This results in “information poverty,”
which is also known as the “gendered digital divide”
(Kennedy et al., 2003). Participation divide studies suggest
that women are significantly less likely to share content on
the web than men (Hargittai & Walejko, 2008). Ono and
Zavodny (2007) illustrate how digital inequality across coun-
tries mirrors existing gender inequality in those countries. For
instance, gender gaps in information technology (IT) use in
countries, such as South Korea, exceed the gaps in more gen-
der-egalitarian countries, such as Sweden and Denmark. This
pattern highlights the important role of social and economic
macro-structures in generating and sustaining digital inequal-
ity in terms of gender and other socio-demographic attributes.
Recent research on gender digital inequality has developed to
include more nuanced analyses that discover emerging topics
related to digital activities, such as differences in offending
behavior (Robinson et al., 2020).
Social media platforms, such as Facebook, are considered
useful arenas for closing potential divides across countries
(Ali, 2011; Xenos et al., 2014). Yet, when we zoom in on
1064906SMSXXX10.1177/20563051211064906Social Media <span class="symbol" cstyle="Mathematical">+</span> SocietyKim et al.
research-article20212021
1Aarhus University, Denmark
2YeungNam University, South Korea
Corresponding Author:
Anja Bechmann, DATALAB, Center for Digital Social Research,
Department of Media and Journalism Studies, School of Communication
and Culture, Aarhus University, Helsingforsgade 14, 8200 Aarhus N,
Denmark.
Email: anjabechmann@cc.au.dk
Systemic Gendering in Facebook
Group Participation
Jiyoung Ydun Kim1, Riccardo Fusaroli1,
Han Woo Park2, and Anja Bechmann1
Abstract
Communication is increasingly taking place in Facebook Groups around the world. Yet, we have little scientific knowledge
of Facebook Groups at scale, especially the extent to which general systemic gendering is a pattern in participation in such
groups. This knowledge deficit is problematic for digitalized and data-driven democratic societies. Therefore, this article
aims to investigate gender differences in open, closed, and secret Facebook Groups. The study relies on a unique large-scale
Facebook Group dataset from a sample that reflects the gender of Facebook users and the Facebook Groups they belong
to in both Denmark and South Korea. By applying Bayesian models and developing a notion of participation that consists
of both structural and actual participation, the study finds that the relation between country, gender, and participation is
strongly modulated by gender differences. Females are more engaged than males in Denmark, while the opposite is true for
South Korea. In both countries, privacy affects females’ participation more than males’. This article contributes to the field
by presenting new large-scale findings that explore gender differences on three levels of Facebook Group privacy settings
(open, closed, and secret) in a hitherto understudied communication space and, by doing so, it highlights the importance of
privacy and country in predicting systemic gendering.
Keywords
participation, network size, open–closed–secret Facebook Groups, gender inequality, privacy, BRMS
2 Social Media + Society
gender, Brandtzaeg’s (2017) Facebook page “likes” study
shows that, instead of closing gender divides in participation,
Facebook in fact reinforces traditional divides, for instance
concerning the kinds of topics users like. It is thus crucial to
investigate whether we see such differences in both Facebook
Group participation and civic participation dimensions
across different countries with different cultural approaches
to gender roles.
In this study, we investigate gender disparities in partici-
pation in Facebook Groups across two countries, South
Korea and Denmark. Although cross-national comparative
studies are always difficult, especially with large-scale data,
we have sampled the two countries to represent nations out-
side the United States with similar internet and smartphone
penetration and a strong Facebook presence, but with very
different approaches to gender gaps on two different conti-
nents: Asia and Europe. We wish to broaden our understand-
ing of how different genders participate and to examine how
this participation is affected by privacy settings, personal
network size, and number of group memberships. We also
wish to investigate actual participation by examining spe-
cific posts and comments in the Facebook Groups. This par-
ticipation approach adds to previously studied civic
engagement forms, such as the gender effects of page likes
and associated topics (Brandtzaeg, 2017). By looking more
broadly at different types of participation in Facebook
Groups across sampled countries, our study aims to contrib-
ute a more nuanced understanding of how country and pri-
vacy may play equally significant roles when examining
gender differences in participation. Although we recognize
all gender identities, for reasons of comparison and limita-
tions in data structure and classifiers, we treat gender as
binary in this article.
In the next section, we will present the theoretical frame-
work, central concepts and literature that helped us form and
support our hypotheses within this research area. We will fol-
low this with a data and methods section in which we justify
our research design, before presenting our results, conclu-
sion and discussion, in which we explore the implications
and limitations of the study and possible avenues for future
research.
Theory and Hypotheses
Defining Systemic Gendering and Participation on
Facebook
Systemic gendering in online participation has often been
described in various ways using similar terms, such as
inequality, differences, and systemic biases that refer to mea-
surable structural differences in participation and engage-
ment patterns: “differences in the online creation and sharing
of purpose-driven content with specific audiences”
(Hoffmann et al., 2015, p. 699). In this context, an existing
study reveals systemic gendering in various aspects of online
communication, in which men participate more than women
in various ways, from the time they spend online to their
level of online political activity (Calenda & Meijer, 2009).
Our notion of systemic gendering thus points to gender dif-
ferences in Facebook Group participation.
Civic engagement is defined as “the degree to which peo-
ple become involved in their community, both actively and
passively, including such political and organizational activi-
ties as political rallies, book and sports clubs” (Quan-Haase
et al., 2002). Importantly, political spaces and apolitical
spaces are not mutually exclusive. In Facebook Groups, peo-
ple interact around public content. Conversations related to
hobbies can initiate discussions on important social or politi-
cal issues (Lee & Park, 2016; Shirky, 2011; Wright, 2012).
Facebook as a communication space is also used by activists
to contour participation in an online territory (Warren et al.,
2014). Brandtzaeg (2012) defines civic engagement on
Facebook as “action in response to societal needs, in the
form of supportive, deliberative, and collaborative practices
in social media” (p. 67). However, he only focuses on “likes”
in his study. Inspired by van Dijck (2009) and Lutz and
Hoffmann (2017) among others, we wish to continue the
notion of online civic engagement found in the work of
Quan-Haase and Wellman (2002) by broadening Brandtzaeg’s
analysis of how civic engagement can take place on
Facebook. We distinguish between two levels of online par-
ticipation: structural participation, in which, by connecting
with friends and joining a group, users secure a level of
information exposure that is relevant to address to discuss
information poverty as a personal network structure; and
actual participation, in which users post and comment and
which reveals the ways frequent users participate in the com-
munication taking place.
Gender and Three Levels of Facebook Group
Privacy
Facebook Groups are increasing in popularity, and not just
among users; the Facebook newsfeed algorithm has changed
to give priority to groups over Facebook pages, which makes
communities as central as friends (Facebook for Developers—
F8 2019 Day 1 Keynote, n.d.). The increasing number of
closed online communities in the social media environment
presents a challenge that requires a range of new research
approaches (Boccia Artieri et al., 2021). By communicating
in a group on Facebook, people can discuss their common
interests in groups that contain either a smaller or larger
number of people than in their newsfeed, and, in this way,
being part of an interest group can expand a person’s net-
work beyond their list of friends.
Due to issues of accessibility, large-scale Facebook
Groups studies tend to focus on the content of particular
open and closed groups (Fernandes et al., 2010). Both quali-
tative and computational studies have been conducted in an
attempt to understand the communication on pages and in
Kim et al. 3
public groups with political, health and technological affor-
dances for education-related content (Bender et al., 2011;
Wang et al., 2012). Since they are more accessible, public
groups are currently the most researched groups in existing
literature. However, there are two other types of groups,
namely, closed and secret groups, with stricter access poli-
cies. Closed groups are visible in the search function, but
require membership to access the group content. Secret
groups do not appear in the search function and members
have to be invited to join (they cannot simply request access;
Facebook Help Center, n.d.).1
Communication privacy management (CPM) theory
explains how people make decisions about revealing private
information to others according to gender and cultural fac-
tors (Petronio, 2012). Based on this strand of literature, we
expect that group privacy settings will play a role when it
comes to the participation of different genders (Cirucci,
2017; Villebro et al., 2018). Women on social media show
higher privacy concerns and participation than men (De
Wolf et al., 2014; Hoy & Milne, 2010; Tifferet, 2019;
Tufekci, 2008). An interesting study on private Facebook
Groups shows that women use female-only closed and
secret Facebook Groups for professional support, network-
ing opportunities, and career development (Pruchniewska,
2019).
The Social Web Gendered Privacy Model (SWGPM;
Thelwall, 2011) also explains gender disparities in privacy
concerns and practices on social media. SWGPM shows that,
owing to four factors—physical security, harassment, social
communication skills, and social communication needs—
females have more of these privacy concerns when using
social media. Privacy settings can give the user a degree of
control over who can read a specific content piece
(Nissenbaum, 2009), and females posting videos of them-
selves could make themselves vulnerable to sexist comments
and even personal violence (Burgess & Green, 2009, pp.
96–97). In the next section, we will operationalize our con-
cept of participation further.
Structural Participation and Actual Participation
in Facebook Groups
Information exposure is essential to determine whether and
how systemic gendering takes place when it comes to infor-
mation poverty. To secure a broad information base, users can
connect with many friends and be members of different
groups with different privacy settings, which in turn give rise
to possible communication from many other users. Brandtzaeg
(2012) studies how page likes play a significant role in the
information poverty challenge, but, in our study, we wish to
emphasize not only a user’s personal network size and num-
ber of group memberships, but also the number of posts and
comments in a group, that is, a user’s actual participation, as
additional elements in the specific challenge of information
poverty and potentially unequal gender patterns.
A fundamental aspect of information exposure and net-
work capital in the networked society (Castells, 2011, 2013)
is the size of a user’s network. The larger the network, the
more diverse information the user is exposed to (Nahon &
Hemsley, 2013). This provides a solid ground for avoiding
information poverty and we will therefore use network size
as an important starting point when examining differences—
working on the assumption that the greater the network, the
larger the information exposure. The size of the network may
influence the strength of the relationship. In small networks,
the bonds are tight, whereas, in large networks, the bonds
tend to be looser (Arnaboldi et al., 2013). A study of Facebook
(Bakshy et al., 2012) confirmed that weak ties are more opti-
mal for spreading new information than stronger bonds of
homogeneity. The bridging social capital potentials of weak
ties on Facebook lie in the site’s technical features, which
lower the cost of maintaining and communicating with a
larger network.
Information exposure on Facebook also depends on a
user’s friend total, group memberships and “like” history,
since the timeline displays all the user’s friends’ posts and
posts from groups the user previously liked. Therefore, we
also consider the number of group memberships as structural
participation. We investigate group memberships by zoom-
ing in on the different privacy setting in groups (open, closed,
and secret) as potential arenas for different kinds of online
participation (Karimi et al., 2014), in which users have the
opportunity to be exposed to information from topic- or net-
work-related publics with stronger or weaker bonds.
This study focuses not only on the structural participation
level, but also on the actual participation level using posts
and comments in Facebook Groups. Rather than making
friends or joining a group for structural participation, the act
of writing a post and leaving a comment in a group indicates
more substantial participation. Posts and comments are two
different types of participation on social media, but they both
represent a more active form of participation than simply
viewing content and liking posts (Aldous et al., 2019). When
posting in the Facebook Groups, the user provides informa-
tion and news, raises issues, or organizes events. Here the
user is an initiator. On the other hand, when the users com-
ment, they rather respond to, support (Hale et al., 2020) or
engage in an existing frame set by the posts. Here the user is
a respondent. Since the act of leaving a post and a comment
represents different participation types in Facebook Groups,
we expect that there will be differences in the way males and
females enact these kinds of participation in groups with dif-
ferent privacy settings. Due to this difference in actual par-
ticipation, we choose to test posts and comments separately.
A Cross-Country Comparative Study
When studying potential gender differences in digital partici-
pation, it is important to account for potential international dif-
ferences, for instance, being a less gender-egalitarian country
4 Social Media + Society
and a more gender-egalitarian country (in our case, South
Korea and Denmark, respectively). We chose to include South
Korea and Denmark since both countries have a high internet
and Facebook penetration, but are ranked very differently on
the Global Gender Gap Index (GGGI; World Economic
Forum, 2019), as illustrated in Table 1. According to the
GGGI, we consider Denmark (ranked 14 out of 153), a gen-
der-egalitarian country, and South Korea (ranked 108 out of
153), a country with a larger gender gap. The GGGI examines
the gap between males and females across four main catego-
ries: economic participation and opportunity, educational
attainment, health and survival, and political empowerment.
We intend to compare Facebook participation in Denmark and
South Korea with a particular focus on gender differences.
Facebook is the most popular social media in Denmark,
and Facebook usage among the Danish population is increas-
ing year-on-year (Danmarks Statistik, 2020). Facebook is the
social media app in South Korean where people across all
generations spent most time. It was found that South Korean
users spent a total of 4.6 billion minutes on Facebook per
month in 2019, followed by Instagram, Naver Cafe, Band,
and Kakao Story. While men prefer open social media, such
as Facebook and Twitter, women prefer to use relatively
closed social media, such as Kakao Story and Band (KISDI,
2019). Compared with women in Denmark, the online envi-
ronment for women in South Korea has more limitations
regarding gender-based discrimination and harassment
(Freedom House, 2019).
Our understanding of cultural differences is inspired by
Hofstede et al. (2010). According to Hofstede, South Korea is
considered a collectivistic society with a score of 18 out of 100.
Korean society has a long-term commitment to the “group,” be
that a family, extended family, or extended relationships.
Denmark is considered an individualist society with a score of
74 out of 100. Danish society has a loosely knit social frame-
work in which individuals are expected to take care of them-
selves and their immediate families only. In general, people in
a collectivistic culture value in-bound relationships, but, para-
doxically, they are more likely to disclose their personal infor-
mation on social media due to a low prior concern for online
privacy (Cho & Park, 2013). There is a lack of research on the
use of Facebook Groups in collectivistic and individualistic
cultures, but it has been shown that gender differences in two
different cultures can have a significant impact on Facebook
users’ networking participation (Oliveira et al., 2016). Men
tend to have more connections to boast a better online pres-
ence, and it may also be easier for male users with exhibition-
istic tendencies in a collectivistic culture to take advantage of
social media visibility compared with female users (Guo et al.,
2018). In Danish culture, a study (Rossi et al., 2016) finds that,
as part of an individualist society, many Danes view their com-
munication on Facebook as private. They mainly communicate
with close friends and family and Danes do not actively partici-
pate in political debates on Facebook with people outside their
friendship group.
There is a lack of studies about the use of Facebook
Groups in collectivistic and individualistic cultures, but gen-
der differences in two different cultures can have significant
impact on Facebook users’ networking participation (Oliveira
et al., 2016). This study will examine gender differences in
three different Facebook Groups with different privacy set-
tings in two countries. We cannot make any assumptions
about Facebook Group participation in different countries on
this basis, but we will return to these studies in our discus-
sion of our results.
Based on the literature reviewed in the previous para-
graphs, we developed the following hypotheses for the rela-
tionship between gender differences and the other factors in
online participation:
H1. We hypothesize gender differences to be reflected in
network size as structural participation. In other words,
we hypothesize that male users will have larger networks
than female users and that this difference will be larger in
South Korea.
H2. We hypothesize gender differences to be reflected in
belonging to groups, as measured by the number of group
memberships as another type of structural participation.
In other words, we hypothesize that male users will have
more group memberships than female users and that this
difference will be larger in South Korea than in Denmark.
We also expect these effects to be modulated by the social
context (group privacy settings), with female users prefer-
ring more private groups (closed and secret ones).
H3. We hypothesize that male users will have more posts
than female users as actual participation and that this dif-
ference will be larger in South Korea than in Denmark.
We also expect these effects to be modulated by the social
context (group privacy settings), with female users prefer-
ring more private groups (closed and secret ones).
H4. We hypothesize that male users will have more com-
ments than female users as actual participation and that this
difference will be larger in South Korea than in Denmark.
We also expect these effects to be modulated by the social
context (group privacy settings), with female users prefer-
ring more private groups (closed and secret ones).
Table 1. GGGI: South Korea and Denmark.
GGGI Rank/value DK SK
Global index Rank 14 108
Value 0.782 0.672
Economic participation
and opportunity
Rank 41 127
Value 0.735 0.555
Political empowerment Rank 17 79
Value 0.421 0.179
GGGI = Global Gender Gap Index; DK = Denmark; SK = South Korea.
Kim et al. 5
Method
Dataset
To conduct this research, we collected data using the Digital
Footprints software (http://digitalfootprints.dk/) developed
by DATALAB at Aarhus University (Bechmann & Vahlstrup,
2015). Digital Footprints is a data-extraction software that
allows researchers to extract data from Facebook, public
streams as well as private data with user consent. A quota
sampling approach was used to mirror the gender, age, educa-
tion, and area of residence of the Facebook population
(Bechmann, 2019). A total of 1,121 South Korean and 1,000
Danish participants were retrieved, including data from the
20,783 groups they were members of at that particular time.
We collected profile data and retrieved posts, comments and
metadata from all the groups, including closed and secret
groups. The data were collected between 1 April 2014 and 30
April 2015, and we collected data back in time for all groups
in the sample. A total of 10,662,832 posts and 44,732,331
comments were collected from the groups. Due to the relative
lack of data collection time caused by Facebook’s application
programming interface (API) policy change in 2015, South
Korean data collection showed less accurate quota sampling
on other demographics compared with Denmark, yet it
yielded in depth data on gender quota sampling.
Data were retrieved with first-degree informed consent
from participants and permission from both countries; from
the Danish Data Agency and the University Institutional
Review Board in South Korea (Bechmann & Kim, 2020).
Despite obtaining all the required formal and legal clearances,
we understand the important ethical dimension of handling
such data (Bechmann & Kim, 2020), and we therefore only
make highly aggregate claims about the dataset not to violate
the privacy of the participants or the other group members.
However, this dataset is a unique online dataset and allows
our research field to understand potential gendered Facebook
Group participation on a large scale across the three types of
groups (with different privacy settings) and across countries.
Analysis
To test each hypothesis, we built Bayesian Poisson regres-
sion models (log link) with the separate measures of online
participation (number of friends, group membership, posts,
and comments) as count (positive integers) outcomes.
To test Hypothesis 1, we used the number of friends as the
outcome measure and country and gender as predictors (both
main effects and interaction):
logFriendsN = Intercept +1 Country + 2 Gender
+3
()
××
×
ββ
β Gender Country×
To test Hypothesis 2, we used group membership (divided by
privacy level) as the outcome measure, and country, gender,
and privacy level as predictors. This implied that we had
three measures of group membership per user: how many
open, closed, and secret groups each user is a member of.
Therefore, we implemented a multilevel structure of the
model: each user might have a different propensity to join
groups (random intercept by user) and a different response to
levels of privacy (random slope of privacy by user):
logGroupMembership = Intercept+1 Country
+2 Gende
i
()
×
×
β
βrr +3 Gender Country + 4
Privacy + 5 Privacy Gen
i
ββ
β
××
×××dder +6 Privacy
Country + 7 Privacy Gender Coun
β
β
×
××××ttry
Note that the subscript i indicates random or varying effects
by participant, that is, the possibility for single users to devi-
ate from the group level estimate. Given that we observed
many users without any group, we also implemented zero
inflated Poisson regression models, with a separate parame-
ter identifying the rate at which zero memberships occur.
To test Hypotheses 3 and 4, we used the number of posts
and comments, respectively, as outcome measures, and
country, gender, and privacy level as predictors. We further
included an offset to control for the number of groups—at
that level of privacy—the user is a member of and tested
whether zero inflation improved the model. Note that we did
not estimate post/comment propensity for users without any
group membership at that level of privacy: if a user was not
a member of a private group, it did not make sense to esti-
mate their propensity to post in a private group:
logPost/CommentN = Intercept+1 Country
+2 Gender
i
()
×
×
β
β++3 Gender Country + 4
Privacy + 5 Privacy Gend
i
ββ
β
××
×××eer + 6
Privacy Country + 7 Privacy Gender
Count
β
β×× ××
×rry + offset logGroupMembership(
()
All models were implemented in a Bayesian framework using
Bayesian Regression Models using Stan (BRMS) in R. Given
that implementing maximal random effect structure and zero
inflation have been shown to more adequately explain datas-
ets and control for false positives, we opted for a Bayesian
framework. All models included weakly informative priors to
discount extreme parameter values without otherwise affect-
ing the statistical inference. All priors were tested via prior
predictive checks to ensure our expected outcomes were on
the right scale (e.g., that, given the chosen priors, expected
group memberships were contained in a range between 0 and
a few hundred—and not a few million).
To assess whether interactions should be included in the
models, we ran a Bayesian Leave-One-Out Cross-Validation
Information Criterion (LOOIC)-based model selection
(Vehtari et al., 2017). LOOIC is a statistical estimation of
out-of-sample error, that is, how much error the model
would make if it were to predict new data (e.g., the number
of friends of new people, only knowing their country and
6 Social Media + Society
gender). We fitted models with and without interactions and
assessed which model was most likely to generalize to new
data (i.e., fitted the data best, while avoiding estimated over-
fitting), by looking at differences in LOOIC and at whether
these differences were credible (low standard error com-
pared with the size of the difference). In summary, predic-
tors are only included if they improve the estimated
out-of-sample performance of the model, and therefore, all
reported predictors in the results section are important in
explaining the data (contrary to a frequentist framework that
would not rely on model comparison but only on signifi-
cance testing). Furthermore, the effects of the predictors are
assessed by plotting the posterior estimates of the models
and reporting the mean and standard deviations of the out-
come measure in the different conditions.
Results
H1. Gender and Structural Participation: Number
of Friends
When testing Hypothesis 1, we find that Danish females
have more friends than males, while South Korean females
have fewer friends than males, as shown in Table 2.
A Loo-based model comparison indicated that the full
model including an interaction between country and gender
had the lowest estimated out-of-sample error (for details on
the test, see Appendix A). Figure 1 shows that the estimates
indicate that South Korean users tend to have more friends
than Danish users, but that effect is mostly driven by male
users, with female users being more similar in their number
of friends.
H2. Gender and Structural Participation: Group
Memberships
Again, a Loo-based model comparison indicated that the full
model, including the interaction between country, gender,
and privacy had the lowest estimated out-of-sample error
without zero inflation (see Appendix B).
According to Table 3, Danish females have more group
memberships than males across the three groups with different
privacy settings. In South Korea, we see the opposite pattern:
South Korean females have fewer group memberships than
males across the three groups with different privacy settings.
Table 2. Mean and SE of Number of Friends.
Denmark South Korea
Female Male Female Male
M (SE) 261 (0.728) 197 (0.65) 273 (0.677) 417 (0.879)
SE = standard error.
Figure 1. Country and gender effect plot on the number of friends.
Table 3. Mean and SE of Number of Group Memberships.
M (SE) Denmark South Korea
Female Male Female Male
Open 7.51 (9.68) 6.39 (10.26) 3.13 (17.87) 6.90 (25.04)
Closed 10.59 (14.17) 5.29 (7.19) 2.00 (4.25) 3.55 (7.92)
Secret 2.84 (3.51) 1.69 (2.77) 1.05 (3.38) 1.59 (4.60)
SE = standard error.
Kim et al. 7
The estimates in Table 3 also indicate that Danish users
tend to have more group memberships than South Korean
users. However, they also indicate that effect is mostly driven
by female users, with male users being close in their number
of memberships, as shown in Figure 2.
H3. Gender and Actual Participation: Posts
The Loo-based model comparison of H3 indicated that the
model including the interactions between country and privacy
and between gender and privacy had the lowest estimated
out-of-sample error without zero inflation. Including a three-
way interaction between country, gender, and privacy or a
two-way interaction between country and gender would not
improve the model and would increase the risk of overfitting
(leading to a higher rate of false positives, see Appendix C).
According to Table 4, Danish females have more group
posts than males in closed and secret groups but not in open
groups. South Korean females have fewer group posts than
males across privacy groups, but the differences in mean are
smaller in secret groups than open and closed groups.
We see different gender patterns in the two countries.
Figure 3a shows the number of posts by country, privacy, and
gender. Males post more than females in South Korea (males:
M = 6.59, SE = 50.83 vs females: M = 3.08, SE = 25.69), but the
opposite holds true for Denmark (males: M = 11.06,
SE = 51.61, vs females: M = 14.47, SE = 38.08). In addition,
while an increase in privacy involves an increase in activity in
South Korea for both genders, in Denmark, closed groups
show the highest activity.
Figure 3b shows the effect between country and privacy.
For both countries, the general trend is that the higher the pri-
vacy levels of the group, the more users post. Interestingly, the
privacy effect is bigger in Denmark than in South Korea. The
difference among the open and closed groups is much clearer
in Denmark, while the difference between all three levels of
privacy is smaller in South Korea. Figure 3c shows the effect
of gender and privacy. The closed and secret groups have more
posts than open groups for both countries. But females have a
bigger effect on privacy than males. Furthermore, females
have half the number of posts on open groups as males.
H4. Gender and Actual Participation: Comments
Similar to post activities (H3), comment activities also show no
effect between country and gender and the three-way interac-
tions of country, gender, and privacy. LOOIC indicated that the
model including the interactions between country and privacy,
and between gender and privacy had the lowest estimated out-
of-sample error without zero inflation (see Appendix D).
As reported in Table 5, Danish females have more group
comments than males across the three groups. In South Korea,
Figure 2. Group membership and effect on country, gender, and privacy.
Table 4. Mean and SE of Number of Posts.
M (SE) Denmark South Korea
Female Male Female Male
Open 5.35 (18.60) 8.69 (41.66) 0.34 (2.07) 3.95 (24.10)
Closed 23.77 (44.74) 15.78 (62.22) 1.65 (9.52) 6.41 (41.58)
Secret 14.31 (42.88) 8.72 (51.71) 7.24 (43.14) 9.41 (73.73)
SE = standard error.
8 Social Media + Society
we see the opposite pattern: South Korean females have fewer
group comments than males in open and closed groups, but
females have more comments than males in secret groups.
Figure 4a shows the number of group posts by country, pri-
vacy, and gender. Males comment more than females in South
Korea (males: M = 6.59, SE = 50.83 vs females: M = 3.08,
SE = 25.69), but the opposite holds true for Denmark (males:
M = 11.06, SE = 52.61, vs females: M = 14.47, SE = 38.08).
Moreover, an increase in privacy involves an increase in activ-
ity for both genders in South Korea.
Figure 4b shows the effect between country and privacy.
Similar to H3, the privacy effect is bigger in Denmark than in
South Korea, but the difference between the closed and
secret groups is much clearer in South Korea. Figure 4c
shows the effect of gender and privacy settings. The closed
and secret groups have more comments than open groups for
both countries, but females have a bigger effect on privacy
settings than males.
The findings show an even larger difference in online par-
ticipation as a function of gender differences than we hypoth-
esized. We see an inverse of relative gender engagement by
country. Males have much fewer friends than females in
Denmark and, significantly, males have more friends than
females in South Korea (H1).
One exploratory analysis, however, indicates that some
caution is warranted, since young people (under 30) in par-
ticular show different effects, with the Danish gender gap
being largely defined by young users (for detailed accounts
on age tests see Appendix E).
The findings also show a larger difference than expected in
the sense that the difference made by gender in the two coun-
tries is opposite: An interesting result is that post and comment
activities show that privacy effects are more prevalent among
females for both countries with nuanced differences between
posts and comments. While three-way interaction of country,
gender, and privacy had a different impact on structural
Figure 3. Group post and effects on country, gender, and privacy. (a) The number of group posts by country, privacy, and gender.
(b) The effect between the country and privacy. (c) The effect between gender and privacy.
Table 5. Mean and SE of Number of Comments.
M (SE) Denmark South Korea
Female Male Female Male
Open 17.61 (83.19) 17.17 (64.36) 1.38 (8.13) 7.34 (37.50)
Closed 123.63 (384.96) 58.29 (321.63) 4.38 (22.41) 11.79 (55.44)
Secret 81.33 (382.88) 30.34 (146.1) 24.17 (376.07) 16.96 (104.51)
SE = standard error.
Figure 4. Group comments and effects on country, gender, and privacy. (a) The number of group comments by country, privacy, and
gender. (b) The effect between the country and privacy. (c) The effect between gender and privacy.
Kim et al. 9
participation (membership, H2) in the two countries, only
two-way interaction, between country and privacy, and
between gender and privacy, had an impact on the number of
posts (H3) and comments (H4) produced (no country and gen-
der interaction). Danish females have higher participation in
all the group privacy settings than Danish males, but they only
have a significantly lower number of posts than Danish males
in the open group. South Korean females have lower participa-
tion in all the group privacy settings than Korean males, but
they only have a higher number of comments than Korean
males in the secret groups. Denmark has a clearly different
participation gap than South Korea between open groups
(public groups) and other types of groups (closed and secret
groups). However, there is a smaller participation gap between
the three types of Facebook Groups in South Korea (H3 + H4).
Discussion and Conclusion
The overall finding of the study is that gender differences do
matter as systemic gendering is found in Facebook Groups
according to the study presented. These differences are
increasingly important to investigate because we spend more
and more time in online communities, and these forums
could potentially give rise to societal inequality patterns. We
see a genuinely clear inversion in the systemic gendering of
online participation in the two countries supported by CPM
theory (Petronio, 2012). Females are more engaged than
males in Denmark, and the opposite pattern is the case in
South Korea.
In structural participation, South Korean males have an
unusually large number of friends compared with the other
three groups of people (South Korean females, Danish males,
and Danish females). One reason for this could be that males in
collectivistic cultures are expected to be more extravert than
women on social media on social media (Guo et al., 2018). For
South Korean males, the number of friends is only one type of
online participation to show sociality. Danish females, how-
ever, tend to have more group memberships in proportion to
their number of friends. The gender gap between the number of
friends a user has and the number of groups she or her belongs
to shows a significant gender difference between the two coun-
tries (Ono & Zavodny, 2007), despite the fact that Facebook
has been studied as a useful platform for closing potential
divides between countries (Ali, 2011; Xenos et al., 2014). In
the actual participation, posts and comments in Facebook
Groups differed by gender in the two countries, but in different
ways: The study shows that Danish females only have less par-
ticipation than Danish males in the number of posts in the open
groups, and South Korean females have more participation
than Korean males but only in the number of comments in the
secret groups. Despite or perhaps because of these findings, we
suggest that more research needs to be conducted on the effect
of culture on gender roles in social media participation.
Our study contributes to the field of social media studies by
providing the first study across open, closed, and secret groups
on Facebook, using large-scale data and examining gender and
national differences in relation to privacy settings in online
communities. This allows us to investigate whether we see such
differences in Facebook Group participation in countries with
different cultural approaches to gender. After Facebook closed
Facebook API for researchers in 2015, it has become more dif-
ficult to collect data on closed and secret groups using the API.
Notably, our study also broadens the concept of online par-
ticipation on Facebook by distinguishing between two levels of
participation: structural participation and actual participation.
We argued that it was necessary to focus on Facebook Groups
because they allow us to examine other variables for online
participation, such as the number of group memberships. In
both Denmark and South Korea, users tend to join fewer
Facebook Groups the higher the privacy settings are.
Conversely, users tend to post and comment more in more pri-
vate groups in their actual participation. Interestingly, the low-
est actual participation as communication initiator (post) and
respondent (comment) is South Korean females in the open
groups, and the number of posts and comments steadily
increases through the closed and secret groups. A possible rea-
son for South Korean women preferring to join secret groups
could be the relatively toxic nature of South Korea’s overall
online environment (World Economic Forum, 2020). Moreover,
in secret groups, South Korean women comment more than
South Korean males. This difference can potentially be
explained by findings in existing studies that South Korean
females are more concerned with posting and commenting in a
closed setting based on the SWGPM (Thelwall, 2011) than the
other three tested gender groups. To examine this further, we
recommend that future studies apply supplementary methods.
In this article, we focus on testing whether there is systemic
gendering in participation, understood as gender differences in
structural and actual participation, as these dimensions remain
under-explored in existing research but theory suggests that
such systemic gendering might exist. When exploring this, we
uncovered the need to have a stronger cross-cultural approach to
gender, participation and privacy settings. In general, the results
suggest that, the more private the group, the more people engage
in it. This conforms well with more general sociological com-
munication theory, in which people are expected to participate
in the communication instead of just lurking (Bechmann, 2019;
van Dijck, 2009). Females and males showed different relative
structural participation in Denmark and South Korea, but they
had the same relative production of posts and comments across
the two countries. In both countries, females tended to post and
comment on secret groups more than males.
Our study revealed differences in the privacy settings of
online gender groups by country. Denmark and South Korea
display different sensitivities to Facebook Group privacy set-
tings; Denmark has a clear participation gap between open
groups and the other two types of groups (closed and secret
groups), whereas South Korea shows a larger gap between
secret groups and the other two types. This may indicate that
South Koreans use closed groups as a more public communi-
cation space, whereas Danes use closed groups more similarly
to the secret groups. However, while these trends are
10 Social Media + Society
statistically robust, by exploring the data, we can also observe
two divergent patterns from this general picture. First, closed
groups seem to play a special role in Danish user participation,
presenting more memberships and actual participation than
any other group. Second, we find that the gender gap in Danes’
use of open groups is smaller than the other types of groups.
Our study is unable to make claims about countries other
than Denmark and South Korea. However, one dimension that
could play a role for the generalizability of the predictions found
in this article is the general social media landscape and penetra-
tion rate of Facebook’s national competitors. The function of
Facebook Groups and Facebook as an online community might
differ across countries and gender. For example, we do see a
different activity pattern in Facebook Groups in South Korea
compared with Denmark, and there are more South Korean
male users on Facebook than females. This might be the result
of a general tendency for South Korean females to use domestic
social media (KISDI, 2015). Facebook is considered an open
platform in South Korea, which means that people use Facebook
to expand their network and make new connections, rather than
talking to people they already know (Jun, 2015).
Our definitions of the open, closed, and secret groups on
Facebook are adapted solely from Facebook’s own definitions,
but actual use of the groups might indicate that users understand
the group privacy settings differently. This is an area that schol-
ars could explore in future studies. Such future research could
crystallize an inductive approach to privacy modulated by the
actual usage patterns that is able to loop back into the defining
features of, for instance, a secret group. In this study, we only
compared the three types of groups labeled by Facebook’s pri-
vacy settings because there was no immediate applicable classi-
fication from existing studies that we could use to qualify a
different robust privacy understanding. However, network size,
topic and tie strength could potentially play a role in a more
inductive understanding of Facebook Groups’ privacy levels
(e.g., smaller groups with close-tie communication discussing
sensitive topics being more secret). The fact that Danish females
are more active in Facebook Groups than Danish males yet
South Korean males are more active than South Korean females
might be better explained by such future analyses. Examining
how these participation patterns correlate with Facebook Group
topics would shed light on the discussion of systemic gendering,
comparing, for example, a gender-egalitarian country like
Denmark and a less gender-egalitarian country like South Korea.
Despite these limitations and suggestions for future
research, this study has provided a more nuanced under-
standing of Facebook participation patterns. In doing so, it
has contributed to the discussion on digital sociology as
Facebook Groups become increasingly popular worldwide
for various types of communication and systemic inequality
thus becomes a challenge for democratic societies’ focus on
diversity in public engagement and opinion.
Acknowledgements
A special thanks to participants who shared their data for the pur-
pose of research, DATALAB developers Peter Vahlstrup & Ross
Kristensen-McLachlan who helped with data retrieval and data pre-
processing of both Korean and Danish datasets, Kwang-Taek Roh
and In-Ho Jo of Cyber Emotion Research Center who helped with
the Korean data collection and IRB process. Furthermore, we would
like to thank Userneeds for agreeing to use their internet panel for
the recruitment of the Danish participants and anonymous review-
ers for their constructive comments on various stages of this
publication.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect
to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support
for the research, authorship, and/or publication of this article: The
research was funded by Aarhus University Research Foundation,
Grant Number AUFF-E-2015-FLS-8-55. The article processing
charges related to the publication was supported by Aarhus
University, School of Communication and Culture.
ORCID iDs
Jiyoung Ydun Kim https://orcid.org/0000-0001-9504-0715
Anja Bechmann https://orcid.org/0000-0002-5588-5155
Note
1. In 2019, Facebook reduced its three privacy settings to two:
public (previously open) and private (previously both closed
and private). However, inside a private group, there is an
option to change the group, so that, it does not appear in the
search function. When this hidden option is enabled, the only
way to join the group is to be invited, just like the previous
secret group (as described in this article).
References
Aldous, K. K., An, J., & Jansen, B. J. (2019). View, like, com-
ment, post: Analyzing user engagement by topic at 4 levels
across 5 social media platforms for 53 news organizations
[Proceedings]. International AAAI Conference on Web and
Social Media, June 11–14, Münich, Germany.
Ali, A. H. (2011). The power of social media in developing nations:
New tools for closing the global digital divide and beyond.
Harvard Human Rights Journal, 24, 185–219.
Arnaboldi, V., Guazzini, A., & Passarella, A. (2013). Egocentric
online social networks: Analysis of key features and predic-
tion of tie strength in Facebook. Computer Communications,
36(10–11), 1130–1144. https://doi.org/10.1016/j.comcom.
2013.03.003
Bakshy, E., Rosenn, I., Marlow, C., & Adamic, L. (2012). The role
of social networks in information diffusion. http://arxiv.org/
abs/1201.4145
Bechmann, A. (2019). Inequality in posting behaviour over time:
A study of Danish Facebook users. Nordicom Review, 40(s1),
31–49. https://doi.org/10.2478/nor-2019-0012
Bechmann, A., & Kim, J. Y. (2020). Big data. In R. Iphofen (Ed.),
Handbook of research ethics and scientific integrity (pp. 427–
444). Springer. https://doi.org/10.1007/978-3-030-16759-2_18
Kim et al. 11
Bechmann, A., & Vahlstrup, P. B. (2015). Studying Facebook and
Instagram data: The Digital Footprints software. First Monday,
20(12). https://doi.org/10.5210/fm.v20i12.5968
Bender, J. L., Jimenez-Marroquin, M.-C., & Jadad, A. R. (2011).
Seeking support on Facebook: A content analysis of breast
cancer groups. Journal of Medical Internet Research, 13(1),
Article e16. https://doi.org/10.2196/jmir.1560
Boccia Artieri, G., Brilli, S., & Zurovac, E. (2021). Below the
radar: Private groups, locked platforms, and ephemeral con-
tent—Introduction to the special issue. Social Media +
Society, 7(1), Article 2056305121988930. https://doi.org/10.
1177/2056305121988930
Brandtzaeg, P. B. (2012). Social networking sites: Their users
and social implications—A longitudinal study. Journal of
Computer-Mediated Communication, 17(4), 467–488. https://
doi.org/10.1111/j.1083-6101.2012.01580.x
Brandtzaeg, P. B. (2017). Facebook is no “Great equalizer”: A big
data approach to gender differences in civic engagement across
countries. Social Science Computer Review, 35(1), 103–125.
https://doi.org/10.1177/0894439315605806
Burgess, J. E., & Green, J. B. (2009). The entrepreneurial vlogger:
Participatory culture beyond the professional-amateur divide.
In P. Snickars & P. Vonderau (Eds.), The YouTube reader (pp.
89–107). National Library of Sweden/Wallflower Press. http://
www.wallflowerpress.co.uk/product/new-titles/youtube_reader
Calenda, D., & Meijer, A. (2009). Young people, the internet and
political participation. Information, Communication & Society,
12(6), 879–898. https://doi.org/10.1080/13691180802158508
Castells, M. (2011). The rise of the network society. Wiley.
Castells, M. (2013). Communication power. OUP.
Cho, S. E., & Park, H. W. (2013). A qualitative analysis of cross-
cultural new media research: SNS use in Asia and the West.
Quality & Quantity, 47(4), 2319–2330. https://doi.org/10.1007/
s11135-011-9658-z
Cirucci, A. M. (2017). Normative interfaces: Affordances, gender,
and race in Facebook. Social Media + Society, 3(2), Article
2056305117717905. https://doi.org/10.1177/2056305117717905
Danmarks Statistik. (2020). It-anvendelse i befolkningen 2019
[The use of IT in the population 2019]. Copenhagen: Statistics
Denmark, Eng. https://www.dst.dk/Site/Dst/SingleFiles/Get
ArchiveFile.aspx?fi=15038101682&fo=0&ext=kvaldel
De Wolf, R., Willaert, K., & Pierson, J. (2014). Managing privacy
boundaries together: Exploring individual and group privacy
management strategies in Facebook. Computers in Human
Behavior, 35, 444–454. https://doi.org/10.1016/j.chb.2014.
03.010
Facebook for Developers—F8 2019 Day 1 Keynote. (n.d.).
Facebook for Developers. https://developers.facebook.com/
videos/f8-2018/f8-2018-day-1-keynote/
Facebook Help Center. (n.d.). https://www.facebook.com/help/2
20336891328465?helpref=search&sr=1&query=group%20
privacy&search_session_id=3bea5a0edac4c2b40a4afafa420
dec1f
Fernandes, J., Giurcanu, M., Bowers, K. W., & Neely, J. C. (2010).
The writing on the wall: A content analysis of college students’
Facebook Groups for the 2008 presidential election. Mass
Communication and Society, 13(5), 653–675. https://doi.org/1
0.1080/15205436.2010.516865
Fortin-Rittberger, J. (2016). Cross-national gender gaps in politi-
cal knowledge. Political Research Quarterly, 69(3), 391–402.
https://doi.org/10.1177/1065912916642867
Fraile, M. (2014). Do women know less about politics than men?
The gender gap in political knowledge in Europe. Social
Politics: International Studies in Gender, State & Society,
21(2), 261–289. https://doi.org/10.1093/sp/jxu006
Freedom House. (2019). Freedom on the Net 2019. https://freedom-
house.org/sites/default/files/2019-11/11042019_Report_FH_
FOTN_2019_final_Public_Download.pdf
Guo, M., Liu, R.-D., Ding, Y., Hu, B., Zhen, R., Liu, Y., & Jiang,
R. (2018). How are extraversion, exhibitionism, and gender
associated with posting selfies on WeChat friends’ circle in
Chinese teenagers? Personality and Individual Differences,
127, 114–116. https://doi.org/10.1016/j.paid.2018.01.042
Hale, B. J., Collins, R., & Kilgo, D. K. (2020). Posting about
cancer: Predicting social support in Imgur comments. Social
Media + Society, 6(4), Article 2056305120965209. https://doi.
org/10.1177/2056305120965209
Hargittai, E., & Walejko, G. (2008). THE PARTICIPATION
DIVIDE: Content creation and sharing in the digital age 1.
Information, Communication & Society, 11(2), 239–256.
https://doi.org/10.1080/13691180801946150
Hoffmann, C. P., Lutz, C., & Meckel, M. (2015). Content creation
on the Internet: A social cognitive perspective on the partici-
pation divide. Information, Communication & Society, 18(6),
696–716. https://doi.org/10.1080/1369118X.2014.991343
Hofstede, G., Hofstede, G. J., & Minkov, M. (2010). Cultures and
organizations: Software of the mind; Intercultural cooperation
and its importance for survival (Rev. and expanded 3rd ed.).
McGraw-Hill.
Hoy, M. G., & Milne, G. (2010). Gender differences in privacy-
related measures for young adult Facebook users. Journal of
Interactive Advertising, 10(2), 28–45. https://doi.org/10.1080/
15252019.2010.10722168
Jun, B. (2015). The usage motivation of SNS: A comparison of
open type and closed type SNS. https://doi.org/10.17662/
ksdim.2015.11.2.181
Karimi, F., Ramenzoni, V. C., & Holme, P. (2014). Structural
differences between open and direct communication in an
online community. Physica A: Statistical Mechanics and
Its Applications, 414, 263–273. https://doi.org/10.1016/j.
physa.2014.07.037
Kennedy, T., Wellman, B., & Klement, K. (2003). Gendering the
digital divide. IT & Society, 1(5), 149–172.
KISDI. (2019). KISDI (Korea Information Society Development
Institute) STAT report 19(10). https://mediasvr.egentouch.
com/egentouch.media/apiFile.do?action=view&SCHOOL_
ID=1007002&URL_KEY=592561de-611e-481e-a65e-
9b27c4ca1cab
Krook, M. L., & True, J. (2012). Rethinking the life cycles of inter-
national norms: The United Nations and the global promotion of
gender equality. European Journal of International Relations,
18(1), 103–127. https://doi.org/10.1177/1354066110380963
Lee, Y., & Park, H. W. (2016). Taking the “pulse” of public opin-
ion from non-political spaces during elections. Journal of the
Korean Data Analysis Society, 18, 2949–2960.
Lutz, C., & Hoffmann, C. P. (2017). The dark side of online par-
ticipation: Exploring non-, passive and negative participa-
tion. Information, Communication & Society, 20(6), 876–897.
https://doi.org/10.1080/1369118X.2017.1293129
Lutz, C., Hoffmann, C. P., & Meckel, M. (2014). Beyond just
politics: A systematic literature review of online participation.
First Monday, 19(7). https://doi.org/10.5210/fm.v19i7.5260
12 Social Media + Society
Nahon, K., & Hemsley, J. (2013). Going viral. Polity Press.
Nissenbaum, H. (2009). Privacy in context: Technology, policy,
and the integrity of social life. Stanford University Press.
Oliveira, M. J., de Huertas, M. K. Z., & Lin, Z. (2016). Factors driv-
ing young users’ engagement with Facebook: Evidence from
Brazil. Computers in Human Behavior, 54, 54–61. https://doi.
org/10.1016/j.chb.2015.07.038
Ono, H., & Zavodny, M. (2007). Digital inequality: A five country
comparison using microdata. Social Science Research, 36(3),
1135–1155. https://doi.org/10.1016/j.ssresearch.2006.09.001
Petronio, S. (2012). Boundaries of privacy: Dialectics of disclo-
sure. SUNY Press.
Pruchniewska, U. (2019). “A group that’s just women for women”:
Feminist affordances of private Facebook groups for profes-
sionals. New Media & Society, 21(6), 1362–1379. https://doi.
org/10.1177/1461444818822490
Quan-Haase, A., & Wellman, B. (2002). How does the Internet
affect social capital.
Quan-Haase, A., Wellman, B., Witte, J. C., & Hampton, K.
N. (2002). Capitalizing on the Net: Social contact, civic
engagement, and sense of community. In B. Wellman & C.
Haythornthwaite (Eds.), The Internet in everyday life (pp. 289–
324). Blackwell. https://doi.org/10.1002/9780470774298.
ch10
Robinson, L., Schulz, J., Blank, G., Ragnedda, M., Ono, H., Hogan,
B., Mesch, G. S., Cotten, S. R., Kretchmer, S. B., Hale, T. M.,
Drabowicz, T., Yan, P., Wellman, B., Harper, M.-G., Quan-
Haase, A., Dunn, H. S., Casilli, A. A., Tubaro, P., Carvath,
R., . . .Khilnani, A. (2020). Digital inequalities 2.0: Legacy
inequalities in the information age. First Monday, 25(7).
https://doi.org/10.5210/fm.v25i7.10842
Rossi, L., Schwartz, S., & Mahnke, M. (2016). Social media use &
political engagement in Denmark. Report 2016. Copenhagen:
DECIDIS - IT University Copenhagen. https://blogit.itu.
dk/decidis/wp-content/uploads/sites/5/2016/03/Decidis_
report_2016.pdf
Shirky, C. (2011). Cognitive surplus: How technology makes con-
sumers into collaborators (Reprint edition). Penguin.
Thelwall, M. (2011). Privacy and gender in the social web. In S.
Trepte & L. Reinecke (Eds.), Privacy online: Perspectives on
privacy and self-disclosure in the social web (pp. 251–265).
Springer. https://doi.org/10.1007/978-3-642-21521-6_18
Tifferet, S. (2019). Gender differences in privacy tendencies on
social network sites: A meta-analysis. Computers in Human
Behavior, 93, 1–12. https://doi.org/10.1016/j.chb.2018.11.046
Tufekci, Z. (2008). Grooming, Gossip, Facebook and Myspace.
Information, Communication & Society, 11(4), 544–564.
https://doi.org/10.1080/13691180801999050
van Dijck, J. (2009). Users like you? Theorizing agency in user-
generated content. Media, Culture & Society, 31(1), 41–58.
https://doi.org/10.1177/0163443708098245
Vehtari, A., Gelman, A., & Gabry, J. (2017). Practical Bayesian
model evaluation using leave-one-out cross-validation and
WAIC. Statistics and Computing, 27(5), 1413–1432. https://
doi.org/10.1007/s11222-016-9696-4
Villebro, M., Shklovski, I., Rossi, L., & Bjørstorp, A. (2018). Comfortably
numb: Danish teens’ attitudes, towards social media platforms
[Proceedings]. 9th International Conference on Social Media
and Society, 18–20 July, Frederiksberg, Denmark: Copenhagen
Business School. https://doi.org/10.1145/3217804.3217911
Wang, Q., Woo, H. L., Quek, C. L., Yang, Y., & Liu, M. (2012). Using
the Facebook group as a learning management system: An explor-
atory study. British Journal of Educational Technology, 43(3),
428–438. https://doi.org/10.1111/j.1467-8535.2011.01195.x
Warren, A. M., Sulaiman, A., & Jaafar, N. I. (2014). Facebook: The
enabler of online civic engagement for activists. Computers
in Human Behavior, 32, 284–289. https://doi.org/10.1016/j.
chb.2013.12.017
World Economic Forum. (2019). The global gender gap report 2020.
https://www.weforum.org/reports/global-gender-gap-report-
2020
World Economic Forum. (2020). The global gender gap report 2019.
https://www.weforum.org/reports/gender-gap-2020-report-
100-years-pay-equality
Wright, S. (2012). From “third place” to “third space”: Everyday
political talk in non-political online spaces. Javnost—The Public,
19(3), 5–20. https://doi.org/10.1080/13183222.2012.11009088
Xenos, M., Vromen, A., & Loader, B. D. (2014). The great
equalizer? Patterns of social media use and youth political
engagement in three advanced democracies. Information,
Communication & Society, 17(2), 151–167. https://doi.org/10
.1080/1369118X.2013.871318
Author Biographies
Jiyoung Ydun Kim is a PhD fellow at the School of Communication
and Culture and a research member of DATALAB, Center for
Digital Social Research at Aarhus University in Denmark. Her
research interests include social media communication, cultural
studies, and digital methods with big data and network analysis.
Riccardo Fusaroli (PhD, Università di Bologna) is an associate pro-
fessor in Cognitive Science at Aarhus University in Denmark. His
research interests include social interactions, experimental semiot-
ics, and metascience.
Han Woo Park (PhD/MA, State University of New York at Buffalo/
Seoul National University) is a professor at Department of Media &
Communication, Interdisciplinary Graduate Programs of Digital
Convergence Business and East Asian Cultural Studies at
YeungNam University in South Korea. His research interests
include webometric hyperlink network analysis, social media use,
triple helix, online big data, and scientometrics.
Anja Bechmann (PhD, Aarhus University) is a professor at Department
of Media and Journalism Studies and director of DATALAB, Center
for Digital Social Research at Aarhus University in Denmark. Her
research interests include social media and collective behavior, digi-
tal communicative dynamics, computational social science, informa-
tion disorders, AI + data ethics and privacy.
Appendix A
The appendix consists of four Leave-One-Out Cross-
Validation Information Criterion (LOOIC) tables for each
hypothesis and three plots to supplement the analysis and
findings in the article.
Kim et al. 13
LOOIC Result of H1
Models LOOIC SE Difference from the best model
Friends 6,64,278.34 60,408.18 45,060.23, SE = 11,939.49
+ country 6,40,099.14 55,326.68 20,881.02, SE = 7,480.20
+ country+ gender 6,36,550.68 54,120.01 17,332.56, SE = 5,382.59
+ country × gender 6,19,218.12 53,171.99 0
LOOIC = leave-one-out cross-validation information criterion; SE = standard error.
LOOIC of H1 indicated that the full model including an interaction between country and gender is the best model which
has the lowest estimated out-of-sample error.
Appendix B
LOOIC Result of H2
LOOIC of H2 indicated that the three-way interaction of country, gender, and privacy improves the model.
Appendix C
LOOIC Result of H3
Models LOOIC SE Difference from the best model
Memberships 36,880.45 799.91 13,418.91, SE = 725.74
+ privacy 23,658.21 185.94 196.67, SE = 34.75
+ country × privacy 23,522.14 186.43 60.60, SE = 25.63
+ gender × privacy 23,604.65 185.84 143.11, SE = 30.45
+ (country + gender) *privacy 23,512.01 187.77 50.48, SE = 19.34
+ country × gender × privacy 23,461.54 187.13 0
Zero-inflated Poisson models
Memberships 35,365.64 786.55 11,904.10, SE = 715.71
+ privacy 23,787.62 185.71 326.08, SE = 35.71
+ country × privacy 23,659.23 197.69 187.42, SE = 26.88
+ gender × privacy 23,732.26 185.75 270.72, SE = 31.32
+ (country + gender) × privacy 23,601.42 187.36 139.89, SE = 20.30
+ country × gender × privacy 23,563.98 187.09 102.44, SE = 16.11
LOOIC = leave-one-out cross-validation information criterion; SE = standard error.
Models LOOIC SE Difference from the best model
Post 69327.73 3645.50 54819.28, SE = 3584.66
+ privacy 14678.99 190.65 170.53, SE = 27.02
+ country × privacy 14516.52 192.53 8.06, SE = 0.84
+ gender × privacy 14687.34 190.96 178.88, SE = 24.48
+ (country + gender) *privacy 14508.46 192.7 0
+ country × gender × privacy 14540.02 193.79 31.57, SE = 0.79
Zero-inflated Poisson models
Post 57158.13 3414.16 42649.67, SE = 3356.56
+ privacy 15518.61 199.01 1010.15, SE = 37.27
+ country × privacy 15302.55 202.18 794.09, SE = 33.36
+ gender × privacy 15486.99 198.94 978.53, SE = 35.96
+ (country + gender) × privacy 15260.33 201.66 751.87, SE = 32.12
+ country × gender × privacy 15248.07 202.05 739.61, SE = 32.50
LOOIC = leave-one-out cross-validation information criterion; SE = standard error.
LOOIC of H3 shows that “Post + (country + gender) × privacy” is the best model.
14 Social Media + Society
Appendix D
LOOIC Result of H4
Models LOOIC SE Difference from the best model
Comments 2,30,514.99 18,721.33 2,11,679.08, SE = 18,674.05
+ privacy 18,954.8 204.43 118.88, SE = 25.51
+ country × privacy 18,853.62 206.55 17.71, SE = 20.17
+ gender × privacy 18,950.57 204.52 114.65, SE = 24.89
+ (country + gender) × privacy 18,835.92 206.72 0
+ country × gender × privacy 18,860.07 207.28 24.15, SE = 20.68
Zero-inflated Poisson models
Comments 21,4219.42 17,434.86 1,95,383.5, SE = 17,387.49
+ privacy 20,642.52 211.87 1,806.6, SE = 48.64
+ country × privacy 20,336.62 214.17 1,500.7, SE = 45.62
+ gender × privacy 20,640.33 212.07 1,804.42, SE = 48.76
+ (country + gender) × privacy 20,325.08 214.37 1,489.17, SE = 45.19
+ country × gender × privacy 20,247.44 214.02 1,411.53, SE = 43.98
LOOIC = leave-one-out cross-validation information criterion; SE = standard error.
LOOIC of H3 shows “Comment + (country + gender) × privacy” is the best model.
Appendix E
Exploratory Analysis on Age
We conducted an exploratory analysis to determine whether age difference in the dataset of the two countries played a role in
our findings in H1, H2, H3, and H4. We filtered three age groups to double-check the age effect on our results (a) under 30, (b)
between 30 and 49, and (c) over 50.
Exploratory Analysis to Check the Age Difference of H1
Kim et al. 15
Exploratory Analysis to Check the Age Difference of H2
Privacy1 = open group, Privacy2 = closed group, Privacy3 = secret groups
16 Social Media + Society
Exploratory Analysis to Check the Age Difference of H3
Kim et al. 17
Exploratory Analysis to Check the Age Difference of H4
The exploratory tests in Appendix E all showed that we need to exercise a degree of caution, especially in the number of
friends (section “Exploratory analysis to check the age difference of H1” in Appendix E), since young people in particular
show different effects by age, with the Danish gender gap being largely defined by young users.
