Making Friends and Communicating On Facebook
Making Friends and Communicating On Facebook
Making Friends and Communicating On Facebook
Social Networks
journal homepage: www.elsevier.com/locate/socnet
a r t i c l e i n f o a b s t r a c t
Keywords: In this paper, we explore the dynamics of access to social capital on Facebook. Existing approaches to
Facebook network-based social capital measures are adapted to the case of Facebook and applied to the friendship
Social capital and communication data of 438,851 users. These measures are correlated to user data in order to identify
Friendship
advantageous behavior for optimizing the possible access to social capital. We find that the access to social
Communication
capital on Facebook is primarily based on a reasonable amount of active communication. Exaggerated
Posting behavior
Homophily friending and posting behavior can deteriorate the access to social capital. Furthermore, we investigate
which kinds of posts are most advantageous as well as questions of homophily based on social capital.
© 2013 Published by Elsevier B.V.
1. Introduction an increase of the number of weak ties (Donath and Boyd, 2004).
Moreover, Steinfield et al. (2008) found on the basis of a longitudi-
The increasing importance of SNSs for maintaining our social nal analysis that Facebook use led to bridging social capital rather
relations cannot be neglected. Users appreciate the possibility to than the other way around. Burke et al. (2011) pointed out that
communicate in an often asymmetric way with many hundreds of receiving messages from Facebook friends contributes to bridging
“friends” or “followers”. On the one hand, the flood of information social capital, while other activities on Facebook do not. However,
can be challenging, on the other hand, this way to communicate is it is unclear whether bonding social capital is boosted by Facebook
said to reduce the effort to build and use social capital. use (Ellison et al., 2007, 2011; Vitak et al., 2011). Both types of social
What is social capital? Lin (2002) summarized the perceptions capital are not affected by online relations to strangers (Ellison et al.,
of influential scholars in the field by “investment in social rela- 2011). SNSs seem to be most valuable to people with lower self-
tions with expected returns in the marketplace”. These investments esteem, weaker social capabilities, or lower life satisfaction (Burke
and returns occur in, for example, the form of knowledge, trends, et al., 2011; Ellison et al., 2007; Steinfield et al., 2008).
ideas, news, and opinions. Sociological literature distinguishes two The cited articles measured social capital by means of sur-
types of social capital: bridging and bonding social capital (Putnam, veys with questions like “I am interested in what goes on at [my
2001). Bonding social capital refers to benefits arising from close community]” and “[my community] is a good place to be”. Social
relations inside of cohesive groups, while bridging social capital is network research usually follows a sociometric approach: Burt
built between groups. Many recent studies are concerned with the (1995) contributed one of the major social network theories related
question of whether the new communication media arising from to social capital. He argued that individuals could build social capi-
the Internet, and especially from SNSs, increase the two types of tal by bridging missing ties between subnetworks (called structural
social capital. Literature seems to agree that the use of Facebook and holes). By adopting the position of a broker, the actor could access
other social media is positively correlated with bridging social cap- and control the information flow between the two otherwise sepa-
ital (Ellison et al., 2007; Shah et al., 2012; Steinfield et al., 2008) and rated groups, which increases its attractiveness for other actors in
the network. Besides brokerage, social capital can also be deter-
mined by centrality, homophily, and density measures and the
reciprocity of relationships (Monge and Contractor, 2003; Borgatti
Abbreviations: API, application programming interface; app, application (soft- et al., 1998; Burt, 2000; Lin, 1999). In this paper, we will undertake a
ware); SNS, social networking sites.
sociometric approach to measure the possible access to social cap-
∗ Corresponding author at: Vienna University of Economics and Business,
ital on Facebook. We argue that although the number of friends
Welthandelsplatz 1, 1020 Vienna, Austria. Tel.: +49 176 37671750.
E-mail addresses: angela.bohn@gmail.com (A. Bohn), christian.buchta@wu.ac.at might be one ingredient of social capital, the actual communi-
(C. Buchta), kurt.hornik@wu.ac.at (K. Hornik), patrick.mair@wu.ac.at (P. Mair). cation ties allow for much more valuable insights. Therefore, we
develop sociometric measures capturing the access to social capi- • The network closure argument, elaborated by Coleman (1988),
tal adapted to the Facebook communication features. We propose for example, suggests that a strong interlocking of entities in
five hypotheses based on sociological, psychological, and social net- a dense network created social capital. This perception corre-
work theories concerning factors influencing the access to social sponds to bonding social capital. A number of arguments support
capital on Facebook. Furthermore, we use regression models to this viewpoint: first, as the information quality decreases with the
reveal the most important main effects and possible interactions path length, direct connections are most adequate to deliver reli-
between predictors. able information. Second, direct connections favor the creativity
output of a network, which make this organizational structure
adequate for many types of corporate- or leisure-related goals.
2. Relevant communication features on Facebook Third, a dense network is more stable against node removals. That
is, the risk of loosing the investment in relationships is diversi-
The subsequent sections require some knowledge about rel- fied. Forth, nobody can escape the social control and sanctions
evant Facebook functionalities that we briefly review here: In of peers, which builds trust. Network measures related to this
essence, the purpose of Facebook from a user perspective is to approach include degree, density, closeness, and page rank, for
become virtual friends with other users, to communicate with example.
them, and to stay informed about their activities and interests. • Burt (1995) argued that bridging structural holes was a recipe for
Friendships are established when a user sends a friend request to building social capital. This concept is related to bridging social
another user and the latter accepts the request. From this point, capital. The structural holes idea was developed in the context
friends can usually read each others’ “posts”. Posts are unaddressed of economic theory. Burt argues that persons in bridging pos-
text messages, possibly enriched by photos or videos, that can itions could benefit from competitive advantages by accessing
be commented on and “liked” (by clicking a “like”-button). They and controlling the information flow between two otherwise
appear on the users’ “news feeds”, a collection of friends’ posts unconnected groups. While information in cohesive groups is
and notifications of other activities of friends (e.g. when someone usually redundant, bridging positions allow for the reception
changed his/her profile picture). Users can post on their own “walls” of different information. Furthermore, the broker can decide to
or on their friends’. Walls show all posts and notifications related establish further contacts between groups and help both of them
to a certain user (whereas news feeds show posts of all of a user’s to benefit from collaboration. Both parties will attribute those
friends). Users can also tag friends in their posts. This way, the post benefits to the broker. These ideas can easily be transferred to
does not only appear on the user’s wall, but also on the tagged per- the context of friendship networks. Betweenness is commonly
son’s wall. Friends’ privacy settings and filter options set by the user used as a measure for bridging positions.
determine which posts and notifications appear on news feeds and
walls.
The two structures are opposing because they can never coexist
on the same node set. There are no structural holes in a closed net-
3. Measuring the access to social capital on Facebook work. However, both structures can appear on different node sets
in the same network. Everyone’s ego network is probably char-
Social capital consists in the exchange of knowledge, trends, acterized by dense parts representing close family, friends, and
ideas, news, and opinions. As opposed to economic capital (money coworkers as well as by some contacts hardly knowing anyone else
and goods) that is exchanged by means of physical transportation in the network. For this reason we calculate the following meas-
or bank transfers, social capital is exchanged through communica- ures capturing the access to both types of social capital. All those
tion. In a network of bank transfers we can evaluate the financial measures are based on communication networks (where edges
power and stability of institutions based on their structural posi- represent the number of likes and comments), not on friendship
tion. In a network of communication ties we can measure the ability networks.
of members to access (and provide) the various forms of social
capital. This is independent of the question of whether persons • Access to rather bonding-like social capital
make use of their social capital (e.g. whether they actually apply
– Reactions R: the number of likes and comments that ego
for the job they heard about; Lin (1999)). However, the ability to
received on his or her posts. This measure is based on the con-
access social capital is manifest. In contrast, the non-existence of
cept of indegree, a frequently used measure of social capital.
communication ties makes it impossible to access social capital.
We presume that social capital is the larger the more reactions
Naturally, the existence of a communication tie does not nec-
on posts someone receives. Likes and comments persons con-
essarily mean that social capital has been transferred. However,
tributed to their own posts are ignored. The measure of R is
seemingly unimportant chats commonly serve to prepare for the
defined on [0, ∞).
transfer of social capital in the future (Sobel, 2002). Previous stud- • Access to bridging social capital
ies, such as Valenzuela et al. (2009) confirmed that there is social
– Betweenness B is the share of shortest paths (geodesics) going
capital on Facebook in terms of life satisfaction, trust, and partici-
through the focal node: B(i) = (gj,k (i))/(gj,k ) where i is the focal
pation.
node, gj,k is the number of geodesics between node j and node
Social structure is not equal to social capital. Instead, social cap-
k (two of i’s friends) and gj,k (i) is the number of geodesics
ital is rather derived from social structure, as pointed out by Portes
between j to k going through i. B(i) is the higher the more
(1998), or, as formulated by Adler and Kwon (2002), “social capital
structural holes are bridged by i. B is defined on [0, 1].
is the resource available to actors as a function of their location in
the structure of their social relations”. A very similar definition was
proposed by XXXX. Based on this idea, a number of appropriate Sometimes, the reciprocity of ties is used as a dimension of social
measures were summarized by Borgatti et al. (1998) and further capital (see e.g. Monge and Contractor, 2003). The idea is to consider
theoretical considerations were elaborated by Burt (2000) and Lin the direction of ties as clearly asymmetric relations can indicate
(1999, 2002). In summary, social network literature proposed two outstanding popularity or inferior social positions. We subscribe
opposing theories about the structures and measurement of social to this viewpoint and we operationalize it on the basis of ratios of
capital: outgoing and incoming ties.
A. Bohn et al. / Social Networks 37 (2014) 29–41 31
• Reciprocity H1a: The access to social capital improves with the number of
– Reciprocity RC: this measure captures the ratio of incoming friends.
and outgoing ties averaged over all friends. RC is defined on [0,
Furthermore, we assume that the number of communication
∞).
partners serves as a better predictor for the access to social
capital than the number of friends. Therefore we formulate
Hypothesis 1b:
4. Hypotheses H1b: The access to social capital improves with the number of
communication partners.
4.1. Hypothesis 1: the access to social capital improves with the
number of friends
4.2. Hypothesis 2: there is a relationship between the posting
frequency and the access to social capital
In many studies and theoretical frameworks, the number of
friends, or more generally the ego-network size, is used as one
According to Simmel (1908), the depth of today’s differentiated
dimension – or the only dimension – of social capital (Borgatti
friendships depends on the extent to which we are willing to share
et al., 1998; La Due Lake and Huckfeldt, 1998) and related
information about ourselves. Can this be translated to the online
concepts like social attractiveness (Walther et al., 2008), popu-
world into “the more I post, the more social capital I am going to
larity (Cha et al., 2010; Harrigan et al., 2012; Szell and Thurner,
build”? Indeed, literature on SNSs found that self-disclosure favors
2010; Tom Tong et al., 2008), and influence (Trusov et al., 2010;
relational closeness, intimacy, well-being, as well as perceived
Xiong et al., 2012). However, network sizes of several hundreds
bridging and bonding social capital (Ko and Kuo, 2009; Ledbetter
or thousands of friends indicate that the notion of friend was
et al., 2011; Rau et al., 2008; Stutzman et al., 2012) unless the con-
extended and has become somewhat unclear in the context of
tent of posts is highly negative (Forest and Wood, 2012). Forest and
SNSs. As Tom Tong et al. (2008) argues, the notion of online
Wood (2012) also stated that users with low self-esteem had diffi-
friend can comprise different kinds of friendship. While litera-
culties to turn their post into social capital. Accordingly, they post
ture agrees that Facebook favors the transformation of latent to
less frequently (Shim et al., 2008). Furthermore, Christofides et al.
weak ties (Ellison et al., 2011) as well as the maintenance and
(2009) found that the need for popularity favors self-disclosure.
strengthening of weak ties (Donath and Boyd, 2004), it is unclear
In contrast, Burke et al. (2011) found that (incoming) private
whether the number of close friends has increased due to the
messages on Facebook are positively correlated to bridging social
new communication technologies. Dunbar (1992) found that the
capital, while other communication media, especially “broadcast-
number of persons we can maintain stable relationships with
ing” (unaddressed posting) are not. What is more, Schoendienst and
(Dunbar’s number) is between 100 and 230 and usually around
Linh (2011) found no significant relationship between the post-
150 due to limitations of the human brain. This number does
ing frequency and the quantity of feedback on posts. Sibona and
not seem to have increased with the emergence of the Inter-
Walczak (2011) found that respondents were more likely to dis-
net (Dunbar, 2012). In contrast, Donath and Boyd (2004) found
solve an online friendship when someone posts too much. They
that SNSs allow for the maintenance of a larger number of close
conclude that very high posting frequencies are responded to nega-
friends.
tively by friends, as with exaggerated friending behavior (discussed
The vague notion of online friend raises the question of whether
above).
its number is still appropriate when defining social capital. There-
Based on prior studies, it is unclear whether the posting fre-
fore, another approach consists in investigating the influence of
quency influences social capital, and if so, in which direction. We
actual friends on social capital. Ellison et al. (2011) showed on
think that this lack of clarity could be due to the fact that there is
the basis of a linear model that there is a slightly positive rela-
a bell-shaped relationship between posting frequency and social
tionship between the number of “actual friends” (respondents
capital, such that social capital increases up to a certain frequency
stated how many of their Facebook friends were actual friends)
and then decreases beyond this level.
and bridging social capital. However, when the squared num-
ber of actual of friends was used as a predictor, this relationship H2: There is a relationship between the posting frequency and
turned curvilinear with a maximum at around 500 friends. The the access to social capital.
authors concluded that friends responded negatively to exag-
gerated friending behavior. Kleck et al. (2007), Tom Tong et al. 4.3. Hypothesis 3: the access to social capital improves with the
(2008), and Utz (2010) came to similar results when inves- share of intentional posts
tigating the relationship between the number of friends and
popularity. On Facebook, not every message that friends can read on their
The social enhancement theory postulates that persons hav- news feeds was created intentionally by the concerned user. In
ing many offline ties tend to maintain many online relations and contrast, Facebook formulates automatic notifications to inform
vice versa. This viewpoint is supported by, for example, Birnie and friends about a user’s activity (e.g. “[Your friend] changed his/her
Horvath (2002) and Mesch and Talmud (2006). The social compen- profile picture”). We refer to these notifications as automatic posts,
sation theory attributes the high online communication activity of whereas intentional posts are created by the user on purpose, like
certain persons to their lack of offline social relations. This con- status posts.
tradicts the social enhancement theory, however, evidence could Existing literature did not take automatic posts into account but
be found for both of them depending on personal characteristics focused on intentional posts. Barash et al. (2010), for example, cat-
like extraversion, self-esteem, and loneliness (Bonetti et al., 2010; egorized posts based on the impression they made on friends. They
Valkenburg and Peter, 2007; Zywica and Danowski, 2008). In the found that entertaining posts had the highest probability of receiv-
context of this paper, we consider the network size as a possi- ing at least one positive comment and depressing posts had most
ble predictor but not a dimension of access to social capital. In negative comments. Solely Burke et al. (2011) suggested that the
line with prior findings mostly supporting the presumption that social capital on Facebook does not only depend on the intensity
social capital is positively associated with network size we propose of use but also on the features used. However, they did not take
Hypothesis 1a: automatic posts into account.
32 A. Bohn et al. / Social Networks 37 (2014) 29–41
Compared to intentional posts, automatic posts display less evi- 5. Facebook data
dence of commitment to the community. Therefore, we think that
the access to social capital improves with the share of intentional In all scientific projects analyzing Facebook data, data collection
posts. was a major challenge. One frequent approach is to crawl profiles
(Catanese et al., 2012, 2011; Gjoka et al., 2011, 2011, 2011; Kurant
H3: The access to social capital improves with the share of inten-
et al., 2011; Mislove et al., 2007; Viswanath et al., 2009). The dis-
tional posts.
advantage of this method is that only public information can be
retrieved, the code needs to be checked and updated constantly as
4.4. Hypothesis 4: the access to social capital improves with the the structure of web files can change without prior notice, and that
number of outgoing addressed communication ties platforms do not provide any documentation about their web files.
Another possibility is to use surveys (Acquisti and Gross, 2006;
Hypothesis 2 was concerned with the influence of unaddressed Bozkir et al., 2010; Young, 2011) or to ask a group of people for the
posts on the access to social capital, that is, posts on someone’s own permission to access their data (Burke et al., 2011; Golder et al.,
wall. This hypothesis addresses the question of whether addressed 2007). Both methods result in a strong sampling bias and make
communication ties, like comments, likes, and tagging friends in it difficult to acquire large samples. Furthermore, surveys suffer
posts, contribute to the access to social capital. According to the from incorrect and missing data. According to Almquist (2012),
social balance theory (XXXX), actors feel comfortable when their false positive and false negative error rates are around 5–20% when
(positive or negative) ties are reciprocated. Thus, the question is: Facebook ties are to be reported.
“Can I use people’s need for reciprocation to increase my social
capital by liking and commenting their content?” Apart from this
need, mutual ties can also be due to truly reciprocated friendships
5.1. Data collection
and shared interests, of course.
Again, we found that literature has not investigated these ques-
The data used in this paper was retrieved by the Facebook appli-
tions so far. Patterns of user interaction have been examined by
cation myFnetwork (http://www.facebook.com/MyFnetwork)
Golder et al. (2007), Viswanath et al. (2009), and Wilson et al.
from November 11, 2011 to January 3, 2012. When users started
(2009), for example, but authors drew no conclusions about the
this app, it asked them for the permission to access the necessary
formation of social capital.
data. If they agreed, the app called the Facebook API on behalf of
Friends probably feels more involved when addressed person-
users. In return, they received a picture of their network graph.
ally. Therefore, we formulate the following hypothesis:
The sna package (Butts, 2008) for R (R Core Team, 2012) was used
H4: The access to social capital improves with the number of for parts of the app as well as for the analysis. Persons became
outgoing addressed communication ties. aware of the app when they saw this picture on their friends’ news
feeds. Similar data collection approaches were conducted by Nazir
4.5. Hypothesis 5: a user’s access to social capital increases with et al. (2008) and Vajda et al. (2011). The app had 1.3 million users
the access to social capital of his or her friends during the data collection period. Our server and scripts were
dimensioned for much less traffic, as we did not expect this level
Homophily is one of the most central theses in SNA. It states of demand. Due to this high utilization, resource limitations, API
that actors with similar attributes have a higher probability of tie limitations, and high data volumes, the data of only 1712 users
formation than actors with different attributes. Homophily was could be retrieved. The waiting queue was implemented in such
found in Facebook relationships concerning race, gender, age, and a way that the selection of those 1712 users was random: when
nationality (Ugander et al., 2011; Wimmer and Lewis, 2010) and the queue was empty, it was filled with a maximum of five users
also in other online communities (Bisgin et al., 2012; Leskovec whose first interaction with the app (in this session) was no longer
and Horvitz, 2008; Mazur and Richards, 2011; Thelwall, 2009, than 25 min ago. Consequently, the selection of users depended
2010). Likewise, Facebook literature suggests that homophily exists on the download time of previous data sets and the point in time
based on behavioral attributes and personality traits: Ugander when users decided to use the app. This way, users were selected
et al. (2011) found a clear degree assortativity in friend networks. randomly during the entire time of data collection. Thirteen of
That is, users with many friends tend to be tied to equally socia- those 1712 data sets were omitted because the user activated
ble people. Utz (2010) came to the conclusion that the perceived the app more than once, such that the data was collected more
extraversion of target persons was positively correlated with the than once. Five additional data sets could not be used because
perceived extraversion of friends. According to Thelwall (2010), of inconsistent or missing data and one dataset was too large to
users of MySpace tend to make friends with persons who have be processed. This makes a total of 1707 available datasets. Each
similar levels of public emotion expression. In contrast, interest- user’s dataset also contains his/her friends’ data. Therefore the
based homophily seems to play only a secondary role (Bisgin et al., entire dataset consists of network information for about 438,851
2012). persons. The data includes the following network data:
Homophily based on social capital might be caused by two pos-
sible mechanisms: one mechanism might reflect the true nature of
• The friendship network: binary symmetric ego networks includ-
homophily in the sense that two persons with similar personalities
ing the app users (egos), their friends, and the friendship relations
and habits happen to be tied regardless of any strategic intentions.
between them.
In contrast, some users might aim to increase their social capital
• The communication network: weighted asymmetric ego
by carefully allocating their ties and hoping that the social capital
networks including the following relations measured on the
of their friends rubs off on them. Both mechanisms can lead to the
same node sets as the friendship networks:
same network structures.
– Like relations: by clicking a “like” button, Facebook users can
From what we can learn from literature, it seems probable that
value another person’s content (posts, photos, videos, and the
homophily based on social capital can be observed:
like).
H5: A user’s access to social capital increases with the access to – Comment relations: Facebook users can make comments on
social capital of his or her friends. another person’s content.
A. Bohn et al. / Social Networks 37 (2014) 29–41 33
– Post relations: Facebook users can post on the “wall” of another in Poland, then in the Netherlands, then in Japan, thereafter in
person to leave non-private messages. the UK and finally in the US. Overall the dataset is dominated
– Tag relations: Facebook users can tag friends in their posts. by European users.
Edge weights and directions correspond to the sum of likes, com- – Throughout the entire time of data collection, there were
ments, and posts a Facebook user contributed to another user’s around twice as many male than female users.
content or wall.
5.2. Notice on privacy The number of friends is the only clear indicator of access to
social capital visible to Facebook users. However, is it a reliable
App users had to explicitly agree that the app could access measure? The first row of plots in Fig. 1 shows the number of friends
the part of their data described above. They were informed that on the x-axis and the three social capital measures on the y-axis.
their data will be used for scientific purposes only and that it Darker shades of hexagons represent higher densities in the cor-
will not be given away to any third party. Users could easily responding cells. Kendall’s correlation coefficients (Cor) are stated
request the deletion of their data from our database. Facebook on top of each plot together with their p-values. The number of
users have the possibility to keep all apps from accessing their data points is 1707 for betweenness B and the corresponding alpha-
data. level is 0.008 = 0.05/(6 × 1) (Bonferroni-correction). R and RC have
438,851 data points with an alpha level of 0.004 = 0.05/(6 × 2). The
5.3. Data limitations same description applies to all hexagon binning plots in this section.
They were produced by using the hexbin package (Carr et al., 2013).
The retrieved data is subject to the following limitations: The first row of plots in Fig. 1 indicates that there is a slightly
positive relationship between social capital measures and the num-
• The retrieved number of posts is limited to the latest 25 for each ber of friends. However, with significant correlations between 0.1
user or friend. 73% of the 438,851 Facebook users have less than and 0.21, the relationship is not as strong as one could expect. The
25 posts. That is, the data reflects their entire posting history. weak relationship with R is especially surprising as it sounds very
(Users can delete their posts.) The remaining 27% of users have reasonable to assume that the larger the audience the more per-
a median posting frequency of 1.8 posts a day. Consequently, sons are interested in the content. However, in our dataset, 51%
25 posts represent nearly two weeks of their posting activity in of communication ties on average are with the three best friends
terms of median. There might be cases where this time frame (in terms of communication intensity) and 80% are with the 5%
or these particular posts are not representative for the recent best friends. On average, app users communicated with only 19%
posting behavior. However, there should not be any systematic of their friends. This confirms the prevailing opinion that the notion
distortion of social capital measures arising from this choice of of online friend was versatile. It seems that Facebook users main-
limit. As Viswanath et al. (2009) pointed out, individual commu- tain only a relatively small number of regular contacts considering
nication ties can change quickly in Facebook, while the overall the low barrier nature of the communication features and the large
network structure is constant over time. number of latent friends. Of course, the stated numbers neglect the
• The Facebook API respects the users’ privacy settings. That is: passively consumed information that entails “mental contact”.
– When friends disallow apps to access their data, the app could The second row of plots in Fig. 1 reveals another insight by show-
not retrieve it. We estimate that this applied to around 2% of ing average social capital measures over an interval of number of
friends. friends. That is, each horizontal dash indicates the length and posi-
– In the friend networks, edges are missing when friends hide tion of the interval on the x-axis and the average social capital for
their friend list from the app user. this interval on the y-axis. Up to a certain limit of 400–600 friends
• Due to limitations on our part we had to reject users with more (indicated by the grey vertical line), social capital increases linearly
than 1000 friends. Facebook users can have up to 5000 friends. with network size. Beyond this level, social capital measures are
• The sampling procedure caused some selection bias: mixed, even in terms of averages. Different choices of intervals yield
– Due to the word-of-mouth spreading of the app, data from very similar results.
some friends was retrieved more than once if friend networks At the limit of 400–600 friends, we might interpret a separa-
overlapped. The difference of the total number of sampled per- tion of Facebook users into two groups: users of the first group
sons to the number of unique ones is 10,294, which corresponds confirm observations by other researchers stating that exaggerated
to 2.3% of the sample. We kept those doublets because time friending behavior had negative impacts. Our analysis suggests that
differences in the data collection and different privacy sett- exaggeration begins at a level of around 400–600 friends. However,
ings of users allow for different perspectives on some friend’s this reasoning does not hold for all users: social capital scores of the
networks. second group might not be primarily supported by mutual commu-
– The average number of friends in our sample is 256, whereas nication ties, a sign of real friendship, but those users are probably
overall it is 130 according to Facebook. This is due to the fact in some kind of outstanding social position or post exceptionally
that Facebook users with many friends were more likely to interesting content.
be aware of the app. Furthermore, app users posted more and Summing up, the high variability in the data suggests that the
received more reactions on their posts than their friends. This number of friends is no clear indicator of access to social capital.
is a sign of self-selection bias in the sense that that passionately In terms of averages, there is a considerable positive association
active Facebook users tend to show more interest in using the between network size and access to social capital up to around
app. 400–600 friends. That is, up to this level, an additional friend is
– The geographical distribution of app users is probably different probably rather advantageous. Beyond, social capital is determined
from all Facebook users. We observed that the app first diffused by different factors.
34 A. Bohn et al. / Social Networks 37 (2014) 29–41
Betweenness
60
Reciprocity
0.8 0.3
0.6 40
0.2
0.4
0.1 20
0.2
0 0
0
0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000
Number of friends Number of friends Number of friends
0.30
12
200
Average number of reactions
10
Average betweenness
Average reciprocity
150
6 8
100
4
50
2
0.05
0
0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000
Number of friends Number of friends Number of friends
6.2. Hypothesis 1b: the access to social capital improves with the the number of friends. Optimal social capital scores are most likely
number of communication partners at a level of 130 communication partners.
As a large number of latent friends does not necessarily con- 6.3. Hypothesis 2: there is a relationship between the posting
tribute to the access to social capital, we investigate the relationship frequency and the access to social capital
between the number of communication partners and social capital.
The first row of plots in Fig. 2 indicates that this association is Fig. 3 shows the average social capital based on the correspond-
indeed much stronger. Especially betweenness, a measure related ing intervals of number of posts per day. The plots are limited to a
to bridging social capital, clearly increases with the number of com- maximum of 30 posts per day, a frequency that is exceeded only by
munication partners (significant correlation of 0.37). very few users.
The second row of plots in Fig. 2 showing social capital meas- For R and RC, there is a clear increase in average social capital up
ures averaged over intervals of number of communication partners, to 0.09–0.11 posts a day (corresponds to 33–40 posts a year) and
underlines the lower variability compared to the corresponding 0.021–0.028 posts a day (corresponds to 8–10 posts a year), respec-
plot of number of friends (Fig. 1(b)). All three averaged social capital tively. Beyond these levels, the associations turn negative, such that
measures steeply increase up to around 130 communication part- someone posting 2.8–3 times a day, for example, receives just as
ners (indicated by vertical dashed lines). Beyond, the average R and many reactions R on average as someone who posts 0.014–0.021
the average RC become heterogeneous, while the average B steadily times a day (corresponds to 5–8 times a year). Although, the rela-
increases. Different choices of intervals give very similar results. tionship is not bell-shaped in the strict sense, an increase, peak, and
This indicates that a large number of communication partners is decline are obvious. Consequently, in order to maximize access to
a good way to improve access to bridging social capital. Instead, bonding social capital on the basis of posting frequency, around one
access to bonding social capital can deteriorate when the number of post every ten days is optimal. When clearly more posts are sent,
communication partners exceeds 130. Likewise, the average RC can friends apparently begin to disesteem the users’ efforts and reduce
decrease beyond a level of 130 communication partners, although their reactions.
an RC of four (the minimum beyond 130 communication partners) The average RC, that is the ratio of incoming and outgoing arcs
is still very advantageous. averaged over all friends, is larger than one for posting frequencies
This level of 130 is astonishingly close the Dunbar’s number of smaller than 3.2 posts a day. Beyond this level, users send more
150 (Dunbar, 1992). This hypothesis does not consider the con- arcs than they receive, which can be interpreted as a disadvan-
tinuity of communication and Dunbar’s number does not aim to tageous and undesirable situation. Everything below 3.2 posts a
optimize social capital. Still, we might interpret the results in a day is gladly accepted and reflected positively by friends. The rela-
way that they confirm rather than disprove the idea that efforts tionship between the posting frequency and B is partly different:
to exceed a certain number of communication partners are not just as the other two measures, the average B strongly increases
necessarily fruitful. for posting frequencies up to around 0.07 a day (26 posts a year).
Summing up, the number of communication partners is a more After this, B continues to grow, but slowly, until the posting fre-
reliable indicator for the access to social capital on Facebook than quency reaches around seven posts a day. Beyond this level, B
A. Bohn et al. / Social Networks 37 (2014) 29–41 35
Betweenness
60
Reciprocity
0.8 0.3
0.6
40
0.2
0.4
0.1 20
0.2
0 0
0
0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250
Number of communication partners Number of communication partners Number of communication partners
0.25
200
8
0.20
150
Average RC
6
Average R
Average B
0.15
100
4
0.10
50
2
0.05
0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250
Number of communication partners Number of communication partners Number of communication partners
is very unsteady. Consequently, access to bridging social capital 6.4. Hypothesis 3: the access to social capital improves with the
clearly grows with the posting frequency as long as it is not very share of intentional posts
high. Once seven posts a day are exceeded, no definite effect can be
shown. Although averaging yielded a clear picture in the previous sec-
As already observed in the two previous hypotheses, reactions tion, the variability in the data is high. This might suggest that
to exaggerated behavior are mixed or negatively. Exaggeration in the access to social capital does not only depend on the num-
terms of posting frequency begins at least at around seven posts a ber of posts but also on the content. An analysis of texts goes
day, according to our data. beyond the scope of this paper, however Facebook provides a
20 40 60
Average R
0 5 10 15 20 25 30
0.05 0.15 0.25
Average B
0 5 10 15 20 25 30
Average RC
0.5 1.5
0 5 10 15 20 25 30
Number of posts per day
Fig. 3. Average of social capital measures (y-axis) vs. the number of posts per day (detail).
36 A. Bohn et al. / Social Networks 37 (2014) 29–41
useful categorization of types of post. We reclassified them into suggest that there is a slight tendency for friendship homophily
two types of posts: intentional and automatic posts. Intentional based on the access to social capital. Fig. 6, showing the averages of
posts are typed and sent by the user. Automatic posts are notifi- social capital measures, indicates roughly bell-shaped relationships
cations formulated by Facebook to inform friends about the users’ with some “outliers” at very high levels of friends’ social capital.
activities. That is, we observe once again a bisection of the dataset into a first
Correlations between the share of intentional posts and social segment with positive relationships between predictors and social
capital measures are 0.32 (R), 0.05 (B, non-significant), and 0.03 capital measures and a second section with rather negative rela-
(RC), suggesting that R is to some extent influenced by the share tionships or high levels of heterogeneity. We conclude that users
of intentional posts, while the other two measures are not. How- having a not so good or medium access to social capital tend to clus-
ever, Fig. 4, showing averaged social capital scores, reveals a more ter in terms of friendship, while persons with a very good access
distinct picture. There is a surprisingly linear relationship between to social capital have mixed friends, many of which range in lower
the share of intentional posts and the average R. We might con- areas.
clude that the share of intentional posts is one of the most decisive Results from Hypothesis 1b suggested that the size of the com-
factors for the access to bonding social capital. Users who do munication network is a better indicator of the access to social
not communicate as much with others, but rather use the site capital than the friendship network size. Accordingly, homophily
for other, more passive, activities, produce a large share of auto- might be more present in the communication network than in the
matic posts and therefore do not get as many reactions. In our friendship network. Fig. 7 shows the averages of users’ social capital
dataset, each Facebook user created only 17% intentional posts measures (y-axis) over corresponding intervals of their communi-
on average. We recommend keeping an eye on the generation cation partners’ average R (left) and RC (right). (B is not available for
of automatic posts like those stated below, because they attract communication partners.) The plots show almost linearly increas-
only a small number of reactions, while increasing the posting fre- ing relationships. That is, users with a not so good or medium access
quency (which can have a negative effect, as noted in the previous to social capital are friends and communicate with persons yielding
section): similar scores. In contrast, users with a very good access to social
“[Friend A] and [Friend B] are now friends”, capital have mixed friends, but they prefer communicating with
“[Your friend] likes a page: http://. . .”,
other high-performing persons.
“[Your friend] went from [former relationship status] to [new relationship status]. . .”,
“[Your friend] installed Facebook for iPhone”, Summing up, homophily based on the access to social capital
“[Your friend] answered [this person’s] question:. . .” exists in the communication network rather than in the friendship
“[Your friend] started using this app:. . .” network.
Status posts with pictures or videos yield most reactions (23.6 on
average with a standard deviation of 41.4). Photos and videos that 6.7. Regression models for the access to social capital
were not uploaded together with a status message gather only 10.1
and 8.1 reactions, respectively, on average. Shared links (“[Your In order to find possible interactions between variables as well
friend] likes a link:. . .”) perform poorest with an average number as to measure the magnitude of main effects, we formulate neg-
of reactions of 2.7. ative binomial hurdle models for R and linear regression models
In contrast, the type of posts does not seem to influence B: up to for B and RC, respectively. Hurdle models consist of two compo-
a share of 40% of intentional posts B is almost constant and beyond nents: a binomial model estimates the probability for observing
this level it is unsteady. Consequently, bridging social capital is not counts larger than zero and a truncated count model is used for
affected by the type of posts. positive counts. This way, the model accounts for excessive zeros.
Hurdle models were fitted using the hurdle() function from the
6.5. Hypothesis 4: the access to social capital improves with the pscl package (Zeileis et al., 2008) and the lm() function from the R
number of outgoing addressed communication ties base distribution was employed for the linear models.
Possible predictors are the number of friends (NF, Hypothesis
Significant correlations of 0.15 (R), 0.35 (B), and 0.46 (RC) indi- 1a), the number of communication partners (NCP, Hypothesis 1b),
cate that there is a noticeable relationship between the number the posting frequency (PF, Hypothesis 2), the share of intentional
of outgoing ties and the access to social capital (first row of plots posts (SIP, Hypothesis 3), the number of outgoing addressed com-
in Fig. 5). Especially the reciprocity-based social capital can be munication ties (OCT, Hypothesis 4), the average of friends’ social
improved by addressing single persons instead of all friends. The capital in terms of R (FSC), and the average social capital of com-
rather weak effect on R, the measure for the access to bonding social munication partners in terms of R (SCP, Hypothesis 5).
capital, might indicate that close friends may reply using other com- As B and RC are highly skewed, they were transformed using the
munication media. In line with this argumentation, B is affected Box–Cox method (Box and Cox, 1964). The parameter optimizing
much stronger, which suggests that loose contacts tend to reply via a normality shape was chosen according to a maximum likeli-
Facebook. hood estimate. The bcPower() function from the car package (Fox
In previous sections, we observed deteriorating or mixed social and Weisberg, 2010) and the boxcox() function from the MASS
capital measures once certain levels of predictors were exceeded. (Venables and Ripley, 2002) package were used for these trans-
As shown in the second row of plots in Fig. 5, a similar effect ema- formations. As the Box–Cox transformation can only be applied to
nating from the number of addressed communication ties is not as positive values, the number of cases was reduced to 1573.
obvious.
6.7.1. Main effect models
6.6. Hypothesis 5: a user’s access to social capital increases with In order to rank predictors according to their importance, we
the access to social capital of his or her friends first present a hurdle model and two linear models, each including
only main effects.
R is available not only for app users, but also for their friends. The main effect negative binomial hurdle model of R using a bino-
We calculated the average R over all friends of each user in order mial distribution for the zero hurdle component is summarized in
to measure the social capital level of ego-networks. Tables 1 and 2. The zero hurdle component models the probabil-
Significant correlations of 0.11 (R), 0.14 (B), and 0.17 (RC) ity for R > 0. P-values in the last column of Table 1 indicate that
between the average R of friends and users’ social capital measures the number and social capital of communication partners and the
A. Bohn et al. / Social Networks 37 (2014) 29–41 37
0.22
1.3
150
1.2
0.18
Average RC
Average R
Average B
1.1
100
0.14
1.0
50
0.9
0.10
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Share of intentional posts Share of intentional posts Share of intentional posts
Fig. 4. Average of social capital measures (y-axis) vs. the share of intentional posts.
1.2
80
1 0.4
Betweenness
60
Reciprocity
0.8 0.3
0.6
40
0.2
0.4
0.1 20
0.2
0 0
0
0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250
N. of addressed comm. ties N. of addressed comm. ties N. of addressed comm. ties
20
250
0.25
200
15
0.20
Average RC
Average R
Average B
150
10
0.15
100
5
0.10
50
Fig. 5. Social capital vs. the number of outgoing addressed communication ties.
25
0.10 0.15 0.20 0.25
400
Average number of reactions
20
Average betweenness
Average reciprocity
300
15
200
10
100
5
0.05
0
Fig. 6. Averages of social capital measures (y-axis) over the corresponding interval of the average R of friends.
38 A. Bohn et al. / Social Networks 37 (2014) 29–41
40
Average reciprocity
10 20 30
50
0
0
Fig. 7. Average of social capital measures of users (y-axis) over corresponding intervals of average social capital measures of ego’s communication partners.
Table 1 Table 3
Main effect negative binomial hurdle model for R (zero hurdle component). Main effect linear model for the Box–Cox transformed B (R2 = 0.28).
Estimate Std. error z-Value Pr(> |z|) Estimate Std. error t-Value Pr(> |t|)
(Intercept) −0.6367 0.4421 −1.4403 0.1498 (Intercept) −1.4605 0.0168 −87.11 0.0000
NF −0.0005 0.0010 −0.4667 0.6407 NF −0.0003 0.0000 −9.94 0.0000
NCP 0.0431 0.0112 3.8422 0.0001 NCP 0.0039 0.0003 15.36 0.0000
PF −0.0088 0.0131 −0.6713 0.5020 PF −0.0004 0.0006 −0.77 0.4432
SIP 19.4546 2.3029 8.4480 0.0000 SIP −0.0295 0.0282 −1.05 0.2956
OCT −0.0035 0.0025 −1.4170 0.1565 OCT 0.0005 0.0001 5.45 0.0000
FSC −0.0187 0.0094 −1.9806 0.0476 FSC 0.0012 0.0004 3.20 0.0014
SCP 0.0389 0.0104 3.7386 0.0002 SCP −0.0000 0.0003 −0.17 0.8686
Table 2
Main effect negative binomial hurdle model for R (count component). Table 4
Main effect linear model for the Box–Cox transformed RC (R2 = 0.30).
Estimate Std. error z-Value Pr(> |z|)
Estimate Std. error t-Value Pr(> |t|)
(Intercept) 2.8135 0.0798 35.2739 0.0000
NF 0.0001 0.0001 0.6487 0.5165 (Intercept) −0.0662 0.0806 −0.82 0.4114
NCP 0.0114 0.0013 9.0727 0.0000 NF −0.0000 0.0002 −0.25 0.8015
PF −0.0141 0.0023 −6.1046 0.0000 NCP 0.0027 0.0012 2.19 0.0290
SIP 2.4850 0.1427 17.4094 0.0000 PF 0.0006 0.0027 0.22 0.8281
OCT −0.0005 0.0004 −1.2181 0.2232 SIP 0.1735 0.1366 1.27 0.2040
FSC −0.0015 0.0018 −0.8030 0.4220 OCT 0.0068 0.0004 15.75 0.0000
SCP 0.0125 0.0013 9.7610 0.0000 FSC 0.0109 0.0019 5.84 0.0000
Log(theta) 0.4165 0.0360 11.5601 0.0000 SCP −0.0022 0.0012 −1.82 0.0695
A. Bohn et al. / Social Networks 37 (2014) 29–41 39
prefer over Facebook. In this sense, the concept of communication- Donath, J., Boyd, D., 2004. Public displays of connection. BT Technology Journal 22
based social capital could be significantly widened. Still, for many (4), 71–82, http://dx.doi.org/10.1023/B:BTTJ.0000047585.06264.cc.
Dunbar, R., 1992. Neocortex size as a constraint on group size
users, their appearance on Facebook is an important aspect of their in primates. Journal of Human Evolution 22 (6), 469–493,
social lives. The analysis could also be extended by other indicators http://dx.doi.org/10.1016/0047-2484(92)90081-J.
of access to social capital on Facebook. A potentially good predic- Dunbar, R., 2012. Social cognition on the Internet: testing constraints on social
network size. Philosophical Transactions of the Royal Society B 367 (1599),
tor of social capital, the number of “subscribers”, was not available 2192–2201, http://dx.doi.org/10.1098/rstb.2012.0121.
at the time of data collection. Subscriptions are a kind of one-sided Ellison, N., Steinfield, C., Lampe, C., 2007. The benefits of Facebook “friends”:
friendship choices. Subscribers can receive some user’s public posts social capital and college students’ use of online social network sites.
Journal of Computer-Mediated Communication 12 (4), 1143–1168,
on their news feeds without needing his or her approval. For exam-
http://dx.doi.org/10.1111/j. 1083-6101.2007.00367.x.
ple, Mark Zuckerberg has 15 million subscribers. Another good Ellison, N., Steinfield, C., Lampe, C., 2011. Connection strategies: social capital impli-
indicator not yet available could be the direction of friend requests. cations of Facebook-enabled communication practices. New Media & Society 13
(6), 873–892, http://dx.doi.org/10.1177/1461444810385389.
Such a variable could help to separate overly expansive persons
Forest, A., Wood, J., 2012. When social networking is not working—individuals
from those being in great demand. Furthermore, the number of pro- with low self-esteem recognize but do not reap the benefits of
file visits could add valuables clues. Therefore, steady extensions of self-disclosure on Facebook. Psychological Science 23 (3), 295–302,
such a project following the development of the API could lead to http://dx.doi.org/10.1177/0956797611429709.
Fox, J., Weisberg, S., 2010. An R Companion to Applied Regression, 2nd edn. Sage,
even deeper insights into the correlates of social capital in SNSs. It Thousand Oaks, California, ISBN 978-1412975148.
remains a challenge of data collection, because every app cannot Gjoka, M., Butts, C., Kurant, M., Markopoulou, A., 2011. Multigraph sampling of
expect to have 44000 users a day, like myFnetwork had during the online social networks. IEEE Journal on Selected Areas Communication 29 (9),
1893–1905, http://dx.doi.org/10.1109/JSAC.2011.111012.
last two weeks of data collection. Gjoka, M., Kurant, M., Butts, C., Markopoulou, A., 2011. Walking in Facebook: a
case study of unbiased sampling of Facebook. In: IEEE INFOCOM 2010, pp. 1–9,
References http://dx.doi.org/10.1109/INFCOM.2010.5462078.
Gjoka, M., Kurant, M., Butts, C., Markopoulou, A., 2011. Practical recommendations
on crawling online social networks. IEEE Journal on Selected Areas in Commu-
Acquisti, A., Gross, R., 2006. Imagined communities: awareness, information sharing, nications 29 (9), 1872–1892, http://dx.doi.org/10.1109/JSAC.2011.111011.
and privacy on the Facebook. In: Danezis, G., Golle, P. (Eds.), Privacy Enhanc- Golder, S., Wilkinson, D., Huberman, B., 2007. Rhythms of social interaction: messag-
ing Technologies. Lecture Notes in Computer Science, vol. 4258, pp. 36–58, ing within a massive online network. In: Steinfield, C., Pentland, B., Ackerman,
http://dx.doi.org/10.1007/11957454 3. M., Contractor, N. (Eds.), Communities and Technologies 2007. Springer, London,
Adler, P., Kwon, S., 2002. Social capital: Prospects for a new concept. The Academy pp. 41–66, http://dx.doi.org/10.1007/978-1-84628-905-7 3, ISBN 978-1-84628-
of Management Review 27 (1), 17–40, http://www.jstor.org/stable/4134367 904-0.
Almquist, Z., 2012. Random errors in egocentric networks. Social Networks 34 (4), Harrigan, N., Achananuparp, P., Lim, E.-P., 2012. Influentials, novelty,
493–505, http://dx.doi.org/10.1016/j.socnet.2012.03.002. and social contagion: the viral power of average friends, close
Barash, V., Ducheneaut, N., Isaacs, E., Bellotti, V., 2010. Faceplant: impression communities, and old news. Social Networks 34 (4), 470–480,
(mis)management in Facebook status updates. In: Proceedings of the Fourth http://dx.doi.org/10.1016/j.socnet.2012.02.005.
International AAAI Conference on Weblogs and Social Media, pp. 207–210 Kleck, C., Reese, C., Behnken, D., Sundar, S., 2007. The company you keep and the
http://www.aaai.org/ocs/index.php/ICWSM/ICWSM10/paper/view/1465 image you project: Putting your best face forward in online social networks.
Birnie, S., Horvath, P., 2002. Psychological predictors of Internet social Paper presented at the annual meeting of the International Communica-
communication. Journal of Computer-Mediated Communication 7 (4), 0, tion Association, San Francisco. http://citation.allacademic.com/meta/p mla
http://dx.doi.org/10.1111/j.1083-6101.2002.tb00154.x. apa research citation/1/7/2/7/5/pages172756/p.172756-1.php
Bisgin, H., Agarwal, N., Xu, X., 2012. A study of homophily on social media. World Ko, H.-C., Kuo, F.-Y., 2009. Can blogging enhance subjective well-being
Wide Web 15 (2), 213–232, http://dx.doi.org/10.1007/s11280-011-0143-3. through self-disclosure? CyberPsychology & Behavior 12 (1), 75–79,
Bonetti, L., Campbell, M., Gilmore, L., 2010. The relationship of loneliness http://dx.doi.org/10.1089/cpb.2008.016.
and social anxiety with children’s and adolescents’ online communica- Kurant, M., Markopoulou, A., Thiran, P., 2011. Towards unbiased BFS samp-
tion. Cyberpsychology, Behavior, and Social Networking 13 (3), 279–285, ling. IEEE Journal on Selected Areas in Communications 29, 1799–1809,
http://dx.doi.org/10.1089/cyber.2009.0215. http://dx.doi.org/10.1109/JSAC.2011.111005.
Borgatti, S., Jones, C., Everett, M., 1998. Network measures of social capital. Connec- La Due Lake, R., Huckfeldt, R., 1998. Social capital, social networks,
tions 21 (2), 27–36 http://www.insna.org/PDF/Connections/v21/1998 I-2.pdf and political participation. Political Psychology 19 (3), 567–584,
Box, G., Cox, D., 1964. An analysis of transformations. Journal of the Royal Statistical http://dx.doi.org/10.1111/0162-895X.00118.
Society 26 (2), 211–252 http://www.jstor.org/stable/2984418 Ledbetter, A., Mazer, J., DeGroot, J., Meyer, K., Mao, Y., Swafford, B., 2011. Atti-
Bozkir, A.S., Mazman, S.G., Sezer, E.A., 2010. Identification of user patterns in social tudes toward online social connection and self-disclosure as predictors of
networks by data mining techniques: Facebook case. In: IMCW 2010, CCIS 96, Facebook communication and relational closeness. Communication Research
pp. 145–153, http://dx.doi.org/10.1007/978-3-642-16032-5 13. 38 (1), 27–53, http://dx.doi.org/10.1177/0093650210365537.
Burke, M., Kraut, R., Marlow, C., 2011. Social capital on Facebook: differentiating uses Leskovec, J., Horvitz, E., 2008. Planetary-scale views on a large instant-messaging
and users. In: Proceedings of the 2011 Annual Conference on Human Factors in network. In: Proceedings of the 17th international conference on World Wide
Computing Systems, pp. 571–580, http://dx.doi.org/10.1145/1978942.1979023. Web, pp. 915–924, http://dx.doi.org/10.1145/1367497.1367620.
Burt, R., 1995. Structural Holes: The Social Structure of Competition. Harvard Uni- Lin, N., 1999. Building a network theory of social capital. Connections 22 (1), 28–51
versity Press, Cambridge, Massachusetts, ISBN 978-0674843714. http://www.insna.org/PDF/Connections/v22/1999 I-1-4.pdf
Burt, R., 2000. The network structure of social capital. Research in Organizational Lin, N., 2002. Social Capital—A Theory of Social Structure and Action. Cambridge
Behavior 22, 345–423, http://dx.doi.org/10.1016/S0191-3085(00)22009-1. University Press, Cambridge, Massachusetts, ISBN 978-0521521673.
Butts, C., 2008. Social network analysis with sna. Journal of Statistical Software 24 Mazur, E., Richards, L., 2011. Adolescents’ and emerging adults’ social networking
(6), 1–51 http://www.jstatsoft.org/v24/i06 online: homophily or diversity? Journal of Applied Developmental Psychology
Carr, D., 2013. ported by N. Lewin-Koh, M. Maechler hexbin: Hexagonal Binning 32 (4), 180–188, http://dx.doi.org/10.1016/j.appdev.2011.03.001.
Routines. http://CRAN.R-project.org/package=hexbin Mesch, G., Talmud, I., 2006. Online friendship formation, communication chan-
Catanese, S., De Meo, P., Ferrara, E., Fiumara, G., Provetti, A., 2011. Crawling Face- nels, and social closeness. International Journal of Internet Science 1 (1), 29–44
book for social network analysis purposes. In: Proceedings of the International http://www.ijis.net/ijis1 1/ijis1 1 mesch.pdf
Conference on Web Intelligence, Mining and Semantics, WIMS’11, pp. 1–8, Mislove, A., Marcon, M., Gummadi, K., Druschel, P., Bhattacharjee, B., 2007.
http://dx.doi.org/10.1145/1988688.1988749. Measurement and analysis of online social networks. In: Proceedings of
Catanese, S., De Meo, P., Ferrara, E., Fiumara, G., Provetti, A., 2012. Extrac- the 7th ACM SIGCOMM conference on Internet measurement, pp. 29–42,
tion and analysis of Facebook friendship relations. In: Abraham, A. http://dx.doi.org/10.1145/1298306.1298311.
(Ed.), Computational Social Networks. Springer, London, pp. 291–324, Monge, P., Contractor, N., 2003. Theories of Communication Networks. Oxford Uni-
http://dx.doi.org/10.1007/978-1-4471-4054-2 12, ISBN: 978-1447140535. versity Press, New York, ISBN 978-0195160376.
Cha, M., Haddadi, H., Benevenuto, F., Gummadi, K.P., 2010. Measuring Nazir, A., Raza, S., Chuah, C., 2008. Unveiling Facebook: a measurement
user influence in Twitter: the million follower fallacy. In: 4th Inter- study of social network based applications. In: Proceedings of the
national AAAI Conference on Weblogs and Social Media (ICWSM), 8th ACM SIGCOMM Conference on Internet Measurement, pp. 43–56,
http://www.aaai.org/ocs/index.php/ICWSM/ICWSM10/paper/viewPaper/1538 http://dx.doi.org/10.1145/1452520.1452527.
Christofides, E., Muise, A., Desmarais, S., 2009. Information disclosure and Portes, A., 1998. Social capital: Its origins and applications in
control on Facebook: are they two sides of the same coin or two modern sociology. Annual Review of Sociology 24, 1–24,
different processes? CyberPsychology & Behavior 12 (3), 341–345, http://dx.doi.org/10.1146/annurev.soc.24.1.1.
http://dx.doi.org/10.1089/cpb.2008.0226. Putnam, R., 2001. Bowling Alone: The Collapse and Revival of American Community.
Coleman, J., 1988. Social capital in the creation of human capital. American Journal Simon & Schuster, New York, ISBN 978-0743203043.
of Sociology 94, 95–120, URL http://www.jstor.org/stable/2780243.
A. Bohn et al. / Social Networks 37 (2014) 29–41 41
Rau, P.-L., Gao, Q., Ding, Y., 2008. Relationship between the level of intimacy and formation on social network sites. Journal of Computer-Mediated Communica-
lurking in online social network services. Computers in Human Behavior 24 (6), tion 15 (2), 314–335, http://dx.doi.org/10.1111/j.1083-6101.2010.01522.x.
2757–2770, http://dx.doi.org/10.1016/j.chb.2008.04.001. Vajda, P., Ivanov, I., Goldmann, L., Ebrahimi, T., 2011. Social game epit-
R Core Team, 2012. R: A Language and Environment for Statistical ome versus automatic visual analysis. In: ICME’11 Proceedings of the
Computing. R Foundation for Statistical Computing, Vienna, Austria 2011 IEEE International Conference on Multimedia and Expo, pp. 1–6,
http://www.R-project.org/ http://dx.doi.org/10.1109/ICME.2011.6011927.
Schoendienst, V., Linh, D.-X., 2011. Investigating the relationship between number of Valenzuela, S., Park, N., Kee, K., 2009. Is there social capital in a social net-
friends, posting frequency and received feedback on Facebook. In: AMCIS 2011 work site? Facebook use and college students’ life satisfaction, trust, and
Proceedings, p. 461 http://aisel.aisnet.org/amcis2011 submissions/461/ participation. Journal of Computer-Mediated Communication 14 (4), 475–901,
Shah, V., Subramanian, S., Rouis, S., Limayem, M., 2012. A Study http://dx.doi.org/10.1111/j.1083-6101.2009.01474.x.
on the impact of Facebook usage on student’s social capital Valkenburg, P., Peter, J., 2007. Preadolescents’ and adolescents’ online communica-
and academic performance. In: AMCIS 2012 Proceedings, p. 27 tion and their closeness to friends. Developmental Psychology 43 (2), 267–277,
http://aisel.aisnet.org/amcis2012/proceedings/ISEducation/27 http://dx.doi.org/10.1037/0012-1649.43.2.267.
Shim, M., Lee, M., Park, S., 2008. Photograph use on social network sites Venables, W., Ripley, B., 2002. Modern Applied Statistics with S, 4th edn. Springer,
among South Korean college students: the role of public and pri- New York, ISBN 978-0387954578.
vate self-consciousness. CyberPsychology & Behavior 11 (4), 489–493, Viswanath, B., Mislove, A., Cha, M., Gummadi, K., 2009. On the evolution of user
http://dx.doi.org/10.1089/cpb.2007.0104. interaction in Facebook. In: Proceedings of the 2nd ACM Workshop on Online
Sibona, C., Walczak, S., 2011. Unfriending on Facebook: friend request and Social Networks, pp. 37–42, http://dx.doi.org/10.1145/1592665.1592675.
online/offline behavior analysis. In: International Conference on System Sci- Vitak, J., Ellison, N., Steinfield, C., 2011. The ties that bond: re-examining the
ences (HICSS), pp. 1–10, http://dx.doi.org/10.1109/HICSS.2011.467. relationship between Facebook use and bonding social capital. In: 44th
Simmel, G., 1908. Untersuchungen über die Formen der Vergesellschaftung. Duncker Hawaii International Conference on System Sciences (HICSS), pp. 1–10,
& Humblot, Berlin, Germany. http://dx.doi.org/10.1109/HICSS.2011.435.
Sobel, J., 2002. Can we trust social capital? Journal of Economic Literature 40 (1), Walther, J., Van Der Heide, B., Kim, S., Westerman, D., Tong, S., 2008. The role of
139–154, http://dx.doi.org/10.1257/0022051027001. friends’ appearance and behavior on evaluations of individuals on Facebook:
Steinfield, C., Ellison, N., Lampe, C., 2008. Social capital, self-esteem, and use of online are we known by the company we keep? Human Communication Research 34
social network sites: a longitudinal analysis. Journal of Applied Developmental (1), 28–49, http://dx.doi.org/10.1111/j.1468-2958.2007.00312.x.
Psychology 29 (6), 434–445, http://dx.doi.org/10.1016/j.appdev.2008.07.002. Wilson, C., Boe, B., Sala, A., Puttaswamy, K., Zhao, B., 2009. User interac-
Stutzman, F., Vitak, J., Ellison, N., Gray, R., Cliff Lampe, 2012. Privacy in inter- tions in social networks and their implications. In: Proceedings of the 4th
action: exploring disclosure and social capital in Facebook. In: Proceedings ACM European conference on Computer systems, EuroSys’09, pp. 205–218,
of the Sixth International AAAI Conference on Weblogs and Social Media, http://dx.doi.org/10.1145/1519065.1519089.
http://www.aaai.org/ocs/index.php/ICWSM/ICWSM12/paper/view/4666 Wimmer, A., Lewis, K., 2010. Beyond and below racial homophily: ERG models of
Szell, M., Thurner, S., 2010. Measuring social dynamics in a mas- a friendship network documented on Facebook. American Journal of Sociology
sive multiplayer online game. Social Networks 32 (4), 313–329, 116 (2), 583–642 http://www.princeton.edu/awimmer/WimmerLewis.pdf
http://dx.doi.org/10.1016/j.socnet.2010.06.001. Xiong, Z., Jiang, W., Wang, G., 2012. Evaluating user community influence in online
Thelwall, M., 2009. Homophily in MySpace. Journal of the American social networks. In: Proceedings of the 2012 IEEE 11th International Conference
Society for Information Science and Technology 60 (2), 219–231, on Trust, Security and Privacy in Computing and Communications, TRUST-
http://dx.doi.org/10.1002/asi.20978. COM’12, pp. 640–647, http://dx.doi.org/10.1109/TrustCom.2012.152.
Thelwall, M., 2010. Emotion homophily in social network site messages. First Mon- Young, K., 2011. Social ties, social networks and the Facebook experience.
day 15 (4) http://firstmonday.org/ojs/index.php/fm/article/view/2897/2483 International Journal of Emerging Technologies and Society 9 (1), 20–34
Tom Tong, S., Van Der Heide, B., Langwell, L., Walther, J., 2008. Too much of a good http://www.swinburne.edu.au/hosting/ijets/journal/V9N1/vol9num1-article2.
thing? The relationship between number of friends and interpersonal impres- html
sions on Facebook. Journal of Computer-Mediated Communication 13, 531–549, Zeileis, A., Kleiber, C., Jackman, S., 2008. Regression models for count data in R.
http://dx.doi.org/10.1111/j.1083-6101.2008.00409.x. Journal of Statistical Software 27 (8) http://www.jstatsoft.org/v27/i08/
Trusov, M., Bodapati, A., Bucklin, R., 2010. Determining influential users in Zywica, J., Danowski, J., 2008. The faces of Facebookers: investigating social enhance-
Internet social networks. Journal of Marketing Research 47 (4), 643–658, ment and social compensation hypotheses; predicting Facebook and offline
http://dx.doi.org/10.1509/jmkr.47.4.643. popularity from sociability and self-esteem, and mapping the meanings of pop-
Ugander, J., Karrer, B., Backstrom, L., Marlow, C., 2011. The anatomy of the Facebook ularity with semantic networks. Journal of Computer-Mediated Communication
social graph. CoRR, abs/1111.4503, http://arxiv.org/abs/1111.4503 14 (1), 1–34, http://dx.doi.org/10.1111/j. 1083-6101.2008.01429.x.
Utz, S., 2010. Show me your friends and I will tell you what type of person you are:
how one’s profile, number of friends, and type of friends influence impression