Content uploaded by Yeslam Al-Saggaf
Author content
All content in this area was uploaded by Yeslam Al-Saggaf on Jul 01, 2017
Content may be subject to copyright.
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
Charles Sturt University, Leibniz-Institut für Wissensmedien & University of Tübingen, Leibniz-
Institut für Wissensmedien
Keywords: Negative emotions, Loneliness, Sadness, Retweeting, Number of followers and
friends
Abstract
Do people who express negative feelings (loneliness, sadness) on Twitter gain or lose online
contacts? To answer this question, we tracked the number of followers and followees of people
who tweeted about loneliness or sadness twice; once when they expressed the negative feeling
and a second time five months later. We compared the networks of those users with the networks
of others who either simply retweeted tweets about loneliness/sadness or (re)tweeted about the
corresponding positive feelings. Using these two comparison groups allows us to examine
whether differences in network size are driven by genuinely expressing (vs. retweeting) a
negative emotion. People expressing loneliness in their tweets, as well as people expressing
sadness in their tweets had smaller networks than people expressing feeling loved or happy. This
effect held only for the original tweets, not retweets, and was – in case of sad/happy – stronger
for the followees than the followers. Moreover, we found that people expressing loneliness also
had smaller friends’ networks five months later than the people expressing feeling loved, and
that the networks of the people expressing sadness became even smaller during the following five
months.
1 Introduction
Social media help people to stay in touch with friends and acquaintances and to expand their
networks. However, some scholars worry that online communication just gives people the
illusion of being connected (Turkle, 2012) and that the positive content encountered on many
social media makes people envious, sad and depressed (Grieve, Indian, Witteveen, Anne Tolan,
& Marrington, 2013; Kross et al., 2013). Despite the positivity norm for social media posts
(Reinecke & Trepte, 2014; Utz, 2015b), there are also social media users who vent their negative
emotions and feelings online (Bazarova & Choi, 2014). By doing so, they not only can relieve
negative emotions, but also may gain social support from others. Nonetheless, there is a risk of
losing more followers in the context of Twitter, in which the relationships between each user are
relatively weak. We are therefore interested in finding out whether expressing negative feelings
is related to the number of followers and followees.
To answer this question, we examined the networks of people expressing loneliness and
sadness on Twitter: once at the time of expressing the feeling and a second time five months
later. We also contrasted them with the corresponding positive feelings (feeling loved, happy).
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
Moreover, whereas previous research often excluded retweets in the data cleaning
procedure (see, for example, Kivran-Swaine, Ting, Brubaker, Teodoro and Naaman, 2014), we
consider retweets as an adequate control condition. boyd, Golder & Lotan (2010) argued that
retweeting is a core practice in Twitter and its conventions should be understood; not
overlooked. boyd, Golder & Lotan (2010) study’s findings highlight several motivations for
retweeting including (1) “To amplify or spread tweets to new audiences”; (2) “To make one’s
presence as a listener visible “; (3) “To publicly agree with someone”; (4) “To validate others’
thoughts”; (5) “As an act of friendship, loyalty, or homage by drawing attention, sometimes via a
retweet request”; (6) “To recognize or refer to less popular people or less visible content”; and
(7) “For self-gain, either to gain followers or reciprocity from more visible participants.” Thus,
by holding the content of the tweet constant, we can examine the role of tweeting an experienced
feeling vs. just retweeting the feeling of someone else’s. Our central variable of interest is the
network size, i.e. the number of followers and followees on Twitter.
We collected 11,880 tweets posted in English, including retweets, expressing loneliness,
sadness, feeling loved and happy from an international sample of different Twitter users and
examined the number of followers and friends of the respective Twitter users. We then
performed several quantitative analyses to answer the following research questions:
- Do people who express loneliness and sadness on Twitter have less followers and
followees than (a) people who express feeling loved/happy and (b) people who just retweet
tweets expressing negative feelings?
- Do people who express loneliness and sadness on Twitter show less interaction with
others in tweets?
- Are network differences between lonely/loved and sad/happy people stable over time?
2 Background
2.1 Tweets and network size
Loneliness and sadness are both negative experiences that are related to depression (Joiner Jr,
1997; Krause, 1987). In the literature, different forms of loneliness are distinguished, ranging
from rather mild impatient boredom over desperation to the more enduring and severe forms
depression and self-depreciation (Rubinstein, Shaver, & Peplau, 1979). In offline contexts,
lonely people report smaller networks (Stokes, 1985). Lonely people have also more difficulties
in receiving social support which can decrease their well-being and eventually lead to depression
(Golden et al., 2009; Joiner Jr, 1997). Lonely people also exhibit lower levels of offline self-
disclosure (Chelune, Sultan, & Williams, 1980; Davis & Franzoi, 1986).
With regard to loneliness expressed on Twitter, Kivran-Swaine, Ting, Brubaker, Teodoro
and Naaman (2014) did an in-depth study of tweets containing the phrase “I’m so lonely” and
looked at the temporal aspect (temporary vs. enduring), context and explicit interactivity
(addressing other person). Most tweets were ambiguous with regard to the temporal aspect, but
from the remaining, more (22.5%) indicated transient loneliness than enduring loneliness
(14.8%). However, this study did not include a control group of people not expressing loneliness
and looked more at different forms of loneliness expressed on Twitter than at relationships with
network size. Our aim is to focus on the network sizes of those who explicitly (re)tweeted
loneliness on Twitter.
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
3
With regard to the relationship between expressing loneliness on Twitter and network size,
there are two lines of argumentation possible. It could be that lonely people also disclose less on
online media and consequently do not gain larger networks and social support from these
networks. However, the relative anonymity on social media could also help lonely people to self-
disclose and talk to others. Al-Saggaf & Nielsen (2014) found that people who experience
loneliness on Facebook are also more likely to disclose personal and relationship information
and their address. Previous findings are mixed. Some studies found that shy people benefitted
from online communication (Baker & Oswald, 2010), others found that people with low self-
esteem recognize the benefits of internet use, but do not reap them because their self-disclosure
are more negative and lead to less positive reactions from their network (Forest & Wood, 2012).
Lee et al. (2013) argued that self-disclosure is the mediating mechanism between use of
online media and decreased loneliness. Their study investigated whether loneliness has a
direct/indirect effect on well-being when mediated by self-disclosure. The findings of their study
showed that loneliness positively influences self-disclosure in a way that lonely people rely on
social network sites to compensate for their unsuccessful offline relations. Their study found that
self-disclosure can reduce feelings of loneliness thereby enhancing well-being.
In a similar vein, große Deters & Mehl (2013) showed in an experiment that posting status
updates more frequently reduces loneliness. Thus, there is some evidence that lonely people
might profit from (active) social media use. However, these studies have been conducted on
Facebook or other social network sites, but not on microblogging platforms such as Twitter.
Twitter is different because networks can be asymmetric – when A follows user B, B does not
have to follow user A back. This allows for a further distinction between followers (people who
follow the user) and followees (people the user follows). Moreover, most Facebook friends are
also known by the users offline; on Twitter, a higher proportion of network members are weak
ties or even strangers (Utz, 2015a). Therefore, if a user often expresses negative emotions on
Twitter, those weak ties or strangers may feel annoyed and therefore unfollow that user.
When it comes to the second negative emotion in the focus of our paper, sadness, we know
that in offline contexts, people tend to suppress negative emotions in public (Lin & Qiu, 2012)
especially sadness (Gross, Richards, & John, 2006). On Twitter, which has been associated with
increased expression of emotions (Lin & Qiu, 2012; Bazarova, Choi, Schwanda Sosik, Cosley, &
Whitlock, 2015), Lin & Qiu (2012) found that people with smaller networks are more likely to
express negative emotions relative to positive emotions. We can build on several studies that
focused on depression, the clinical form of sadness, on Twitter. These studies usually focused on
the content of the tweets. For example, Bollen (2015) tracked the tweets of people stating on
Twitter that they were diagnosed with depression, and analyzed the tweets in the months before
and after the tweet about the diagnosis. They also compared the tweets of the depressive Twitter
users with tweets of a random sample of Twitter users and identified the word that were used
more frequently by depressive people. De Choudhury, Counts, & Horvitz (2013) used clinical
depression inventories to determine depression and related the depression scores to linguistic
characteristics of tweets and user and network characteristics. They found that the depressed
users had a smaller numbers of followers and especially followees than the not-depressed users.
The present paper focuses not exclusively on clinical depression, but also on milder and transient
forms of sadness.
Our first research question is therefore:
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
RQ1: Do people who express loneliness or sadness on Twitter have a smaller number of
followers and followees than a) people who express feeling loved or happy b) people who just
retweet tweets expressing the respective feeling?
2.2 Tweets and interactions
Kivran et al. (2014) found that people expressing enduring loneliness were less likely to engage
in interactions on Twitter and consequently also received fewer responses than people expressing
transient loneliness. We will compare the tweets expressing negative feelings with tweets
expressing positive feelings. Based on the findings of Kivran et al. (2014) we expect that people
who express loneliness are less likely to mention another person in this tweet than people
expressing feeling loved or happy.
H1: People who express feeling lonely show less interaction with others than people who
express feeling loved.
It is less clear whether sadness is a directed emotion. Sadness is as loneliness related to
depression, so people expressing sadness could also engage less in interactions. However,
sadness can be caused by many reasons, so feeling sad, at least in a transient way, doesn’t
necessarily imply a smaller network. The saying “a sorrow shared is a sorrow halved” indicates
that talking about sadness might be beneficial, but it is unclear whether people who express
sadness on Twitter prefer talking to all followers or to a specific person.
Our second research question is therefore:
RQ2: Do people who express sadness on Twitter show less or more interaction with others than
a) people who express happiness and b) people who retweet tweets expressing sadness? While
this question pays attention to the interaction within sad/happy tweets, H1 above focuses on the
interaction within lonely/loved tweets.
2.3 Network overtime
The most important question is whether Twitter use helps lonely/sad people to become less
lonely/sad over time. For Facebook, there are findings that self-disclosure on Facebook can help
lonely people to overcome their loneliness (Lee et al., 2013). On Twitter with its asymmetric
networks it is at least easy to follow many people. It might be more difficult to gain followers.
Thus, it is also possible that lonely/sad Twitter users do not fully reap the potential benefits of
getting new connections, for example, after explicitly expressing loneliness/sadness online, they
give others a bad impression and therefore lose connections and contacts on Twitter. We
therefore pose an open research question:
RQ3: Do people who express loneliness/sadness on Twitter have smaller networks several
months later?
3 Methods
3.1 Process of collecting data
This study collected data using the TAGs App and the ‘TwitteR’ and SQLite Packages in R. A
total of 5940 tweets were retrieved on June 9, 2015 from Twitter using the TAGS App after
setting the maximum number of tweets to be returned to 3000 tweets and executing the Run Now
Script, so data can be retrieved once. 2970 tweets were returned after the phrase “I feel lonely”,
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
5
in double quotations marks, was entered in the TAGS search area. Similarly, 2970 tweets were
returned after the phrase “I feel loved”, in double quotations marks, was entered in the TAGS
search area. A further 5940 tweets were retrieved on June 9, 2015 from Twitter by repeating the
same process for the phrases “I am sad” and “I am happy”.
The TAGS App (Version 6.0) is a Google Sheet template developed by Martin Hawksey
(https://tags.hawksey.info/) and made freely available for users to run automated collection of
search results from Twitter. For the TAGS App to work, another App was developed in Twitter
(https://apps.Twitter.com/) to automatically authorise the TAGS App to retrieve data from
Twitter on behalf of the user. After downloading the dataset from Google Sheets it was imported
into SPSS (IBM SPSS Statistics Version 20) for analysis.
The 11,880 tweets are not the only tweets containing these phrases that ever posted to
Twitter; rather those posted to Twitter in the last seven days prior to June 9, 2015. Along with
the tweets, the TAGS App also returned the unique tweet ID, the Twitter user-name, the time the
tweet was posted, the sender’s language, the sender’s unique user ID, the source of the tweet, the
sender’s profile image url, the sender’s followers, the sender’s friends, the sender’s status url, the
hashtags included in the tweet and the ‘in_reply_to_screen_name’.
To examine whether the networks of people tweeting about loneliness/sadness are only
smaller at the moment of tracking the tweets, or whether they also stay smaller over time we
collected information about the number of followers and followees a second time. Using the
‘TwitteR’ package in R and SQLite package for R, we tracked the number of followers and
followees of the people tweeting “I feel lonely”/ “I feel loved” and “I am sad”/ “I am happy” five
months later. We focused only on the people who had sent original tweets and excluded the
people who had retweeted posts about feeling lonely/loved or feeling sad/happy.
We could not obtain information from 395 individuals in the lonely condition and from 458
individuals in the loved condition, leaving us with 1573 and 1993 users (some individuals
appeared repeatedly in the dataset, therefore the number of individuals is lower than the number
of tweets tracked). Thus, roughly 20% in both conditions had used a different Twitter name or
stopped using Twitter altogether. Similarly, we could not obtain information about the number of
followers and followees from 73 users in the sad group and from 69 in the happy category.
3.2 Process of data cleaning and analysis
The purpose of this study was to determine if a relationship exists between expressing the feeling
of loneliness or sadness (vs. positive feelings) and the number of followers and followees. We
used the number of followers and followees as dependent variables in a repeated measurement
analysis of variance, and treated the content of the tweets (lonely, loved, sad, and happy) and the
type of tweet (original tweets vs. retweet) as independent variables. This allows us to detect also
potential interaction effects between type of tweet and content of the tweet as well as diverging
patterns for the number of followers and followees. In line with best practices for reporting
analysis of variance, we will report all significant effects and also unpredicted effects less central
to our research questions. Both dependent variables were continuous; while the independent
variables were nominal variables.
Tweets were manually checked for popular song titles or quotes, see (Kivran-Swaine, Ting,
Brubaker, Teodoro, & Naaman, 2014) for a similar approach; these non-original tweets were
excluded from the analysis (175 in the lonely dataset, 53 in the loved dataset, 21 in the sad
dataset, 49 in the happy dataset). One Twitter account had posted 732 “I am sad” tweets linking
to different pictures and was therefore also excluded from the data analysis. We also excluded
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
two accounts that tweeted “I am happy to help” or something equivalent to help customers
instead of expressing genuine happiness.
Tweets that were retweets were not deleted in this process; instead, a variable RT with the
values 0 for an original tweet and 1 for a retweet was created. Similarly, we created a variable for
interactions when a tweet started with @.
Since both the number of followers and the number of followees were positively skewed
(i.e. did not display a normal distribution), a log-transformation (using LN (number of
followers/followees+1)) was used to normalize the distribution. We added 1 to keep users with 0
followers or followees in the dataset. The analyses were conducted with the log-transformed
values, but to better grasp the results, we reported the backtransformed means.
4 Findings
4.1 Network Size
Lonely and loved
A 2 (type: original tweets vs. retweet) x 2 (content: lonely vs. loved) x 2 (connection type:
number of followers vs. number of followees) analysis of variance with repeated measurements
on the last factor (log-transformed) revealed a significant main effect of type of tweet, F(1,5738)
= 39.50, p < .001, ηp2 = .007, that was qualified by the interaction between type of tweet and
content, F(1,5738) = 17.10, p < .001, ηp2 = .003. In general, people who retweeted a tweet had
larger networks than people who posted an original tweet, but this effect was only significant for
the people tweeting about loneliness; not feeling loved. People who originally tweeted about
loneliness had smaller networks (M = 307) than people retweeting loneliness tweets (M = 478).
There was no significant difference in network size for people tweeting about feeling loved (M =
377) and retweeting tweets about feeling loved (M = 413). Please note that the repeated
measurement analysis collapses for effects not involving connection type across the number of
followers and followees; thus the above reported values refer to these collapsed values. There
was also an interesting main effect of connection type, F(1,5738) = 154.40, p < .001, ηp2 = .03,
indicating that the number of followers in general was higher (M = 428) than the number of
followees (M = 353). No other main or interaction effects were significant, indicating that the
finding on the smaller network size of people tweeting about loneliness holds for followers and
followees. See Table 1 for the means.
Table 1: The means of network size for Lonely and Loved
Original tweet
Retweet
Lonely
Loved
Lonely
Loved
Network size
307
377
478
413
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
7
Sad and happy
Likewise, A 2 (type: original tweets vs. retweet) x 2 (content: sad vs. happy) x 2 (connection
type: number of followers vs. number of followees) analysis of variance with repeated
measurements on the last factor (log-transformed) was conducted. This revealed a significant
main effect of content, F(1,4879) = 9.35, p < .01, ηp2 = .002. Individuals tweeting that they are
sad had smaller networks (M = 320) than individuals tweeting that they are happy (M = 368).
There was also a main effect of connection type, F(1,4879) = 17.28, p < .001, ηp2 = .004,
indicating again that the network of followers (M = 358) is usually larger than the network of
followees (M = 330). However, this effect was qualified by the interaction with type of tweet,
F(1, 4879) = 31.29, p < .001, ηp2 = .006; it was only significant for the original tweets (Ms = 375
vs. 309), but not for retweets (Ms = 341 and 351), (means again collapsed across followers and
followees). The three-way interaction between type of connection, type of tweet and content was
also significant, F(1, 4879) = 5.99, p < .05, ηp2 = .001. The differences in network size between
individuals expressing sadness and individuals expressing happiness were not significant for the
number of followers in original tweets, but were significant for all other comparisons. See Table
2 for the means.
Table 2: The means of network sizes for Sad and Happy
Original tweet
Retweet
Sad
Happy
Sad
Happy
Followers
370
380
308
379
Followees
284
337
324
380
Lonely vs. sad
We ran an additional 2 (type: original tweets vs. retweet) x 2 (content: lonely vs. sad) x 2
(connection type: number of followers vs. number of friends) analysis including only the people
expressing loneliness and the people expressing sadness to see whether these two groups
differed. This time, all effects were significant, so we focus on the most important three-way
interaction, F(1,4909) = 17.52, p < .001, ηp2 = .004, that qualifies all main effects and two-way
interactions. The means correspond to the ones reported in the prior two sections; the direct
comparison now showed that the networks of people expressing loneliness and sadness only
differed significantly in case of the retweets, Fs(1, 4909) = 36.51 and 31.70 for followers and
followeess, respectively, p < .001, ηp2 = .007 and .004, but not for the original tweets, Fs(1,
4909) = 2.09 and 0.26, ns. Thus, both, people expressing loneliness and people expressing
sadness on Twitter, have equally small followees networks. However, people who retweet tweets
about sadness have much smaller networks than people retweeting tweets about loneliness. See
Table 3 for the means.
Table 3: The means of network size for Sad and Lonely
Original tweet
Retweet
Sad
Lonely
Sad
Lonely
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
Followers
370
345
308
502
Followees
284
270
324
429
4.2 Interactions
Ackland (2013) notes that @replies are a better indication of the existence of a social tie between
two users than friends/followers lists. Ackland (2013) argues that there is a problem with
@replies. @replies can be used to simply mention another user, which means it is not necessarily
a direct tweet to them. For this reason Twitter distinguishes between @replies and @mentions. If
the tweet begins with @, it is a @reply; if it does not but includes the @, it is a @mention. Since
@replies are direct messages to users and as Ackland (2013) noted they are a better indication of
the existence of social ties between two users than friends/followers lists, the
‘in_reply_to_screen_name’ column is used.
To compare the percentage of @replies across the four conditions, we focused only on the
original tweets and conducted a cross-table analysis. In line with H1, the χ2(3, N=9926) =
376.46, p < .001 test indicated significant differences between the groups. Only 13.5% of the
tweets expressing loneliness contained @replies, in contrast to 30.9% of the tweets expressing
sadness, 34.7% of the tweets expressing happiness and 38.6% of the tweets expressing feeling
loved. Thus, the positive feelings and sadness are more often expressed in direct interactions,
whereas loneliness is also expressed in a non-directional way. H1 is thereby supported. Table 4
offers a comparison of the percentage of @replies across the four conditions.
Table 4: Summary of the cross-table analysis for the four conditions
Original tweets
Lonely
Loved
Sad
Happy
Percentage of @replies
13.5%
38.6%
30.9%
34.7%
4.3 Longitudinal effects
We wanted to examine whether the networks of people tweeting about feeling lonely/loved and
feeling sad/happy are only smaller at the moment of tracking the tweets, or whether they also
stay smaller over time. Therefore, we tracked the number of followers and followees of the
people tweeting “I feel lonely”/ “I feel loved” and “I feel sad”/ “I feel happy” again after five
months. For more information on how this data was collected see Section 3.1 above.
We conducted a 2 (content) x 2 (connection type) x 2 (time) analysis of variance with
repeated measures on the last two factors. This analysis revealed a theoretically less interesting
main effect of connection type, F(1,3546) = 113.11, p < .001, ηp2 = .03, indicating that the
follower networks were on average larger (M = 381) than the followees networks (M = 322).
More important, the main effect of content was also significant, F(1,3546) = 8.03, p < .01, ηp2 =
.002, indicating again that people tweeting about feeling lonely had smaller networks (M = 331)
than people tweeting about feeling loved (M = 372). No other effect was significant. A few
interactions were marginally significant, but due to the large number of cases we decided not to
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
9
interpret marginally significant interactions. This means, that lonely people still have smaller
networks five months later, and for both groups, network size remained stable.
Similarly, we conducted a 2 (content) x 2 (connection type) x 2 (time) analysis of variance
with repeated measures on the last two factors for the sad/happy dataset. This analysis also
revealed a significant main effect of content, F(1,3054) = 6.12, p < .05, ηp2 = .002. People
tweeting about feeling sad had, averaged across both time points, smaller networks (M = 318)
than people tweeting about feeling happy (M = 364). The main effect of connection type was
also significant, F(1,3054) = 57.70, p < .001, ηp2 = .02, and qualified by the interaction with
content, F(1,3054) = 8.17, p < .01, ηp2 = .003. The general pattern, that the network of followers
is larger than the network of followees, was more pronounced for the people tweeting about
sadness (Ms = 359 vs. 281, respectively, ηp2 = .016) than for the people tweeting about happiness
(Ms = 385 vs. 345, respectively, ηp2 = .004). There was also a significant interaction between
content of the tweets and time, F(1,3054) = 5.97, p < .05, ηp2 = .002. The networks of the people
expressing sadness decreased significantly from 324 to 312, whereas the networks of the people
expressing happiness showed a non-significant increase from 361 to 368. Note that these are the
collapsed values across followers and followees since the three-way interaction was only
marginally significant, F(1,3054) = 3.25, p < .10, ηp2 = .001.
5 Discussion
We compared tweets about two negative feelings (loneliness and sadness) with tweets about the
corresponding positive feelings (feeling loved and happy) as well as with retweets of tweets
containing these expressions and found interesting differences in network size. Our first RQ was
about the network size of people expressing loneliness or sadness. Whereas people expressing
loneliness had less followers and less followees than people expressing feeling loved, people
expressing sadness had only less followees than people expressing happiness.
However, the two groups expressing a negative feeling did not differ from each other in
network size, indicating that loneliness and sadness go both hand in hand with smaller networks.
Interestingly, the effect on network size appeared only for the original tweets (vs. retweets) in
case of loneliness, but for original tweets and retweets in case of sadness. They did differ
however in their level of interactivity; people expressing loneliness were much less likely to
direct their tweets to a specific user than all the other three groups. This is in line with our
hypothesis and the findings by Kivran et al. (2014). The answer to our second RQ, whether
people expressing sadness engage in more or less interactions, is that there are no differences
between people expressing sadness and people expressing happiness or feeling loved. Most
importantly, in answering RQ3, we found that people expressing loneliness on Twitter still had
smaller networks five months later than the people expressing feeling loved. For both groups,
network size remained stable. The picture was different when looking at the people expressing
sadness: Their networks significantly decreased over time. Thus, whereas people expressing
loneliness on Twitter do not at least become lonelier (in terms of Twitter followers and
followees), people expressing sadness on Twitter end up with smaller networks on Twitter over
time. Taken together, these results indicate that the two feelings, although they are often related
and both are related to depressive symptoms, correspond to different patterns. People expressing
loneliness seem to also have smaller networks on Twitter, as expressed in both, smaller number
of followers and followees. At least when they state their loneliness, they address other people
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
less frequently. This is in some way consistent with Al-Saggaf and Nielsen (2014) study’s
findings that fewer lonely Facebook users revealed their status updates compared to those feeling
connected and more lonely users did not disclose their wall compared to the few who did. But
this might be also due to the specific tweet we tracked. Telling another person that one feels
lonely might be perceived as contradiction because it would mean that one has at least one
interaction partner.
People expressing sadness however interacted with other Twitter users in roughly one third
of the tracked cases. Similar interaction rates were found for the positive feelings, indicating that
sad people are not less likely to share their feelings with a specific other person on Twitter than
people who feel happy and loved. This could be because sadness can be caused by external
factors (an accident, loss of a loved person) not in the control of the person. Expressing
loneliness might lead to personality attributions, i.e. the assumption that something is wrong with
the lonely person. However, the networks of the people expressing sadness that were already
smaller at the beginning (especially the number of followees) became even smaller within the
next months, indicating that sadness might be an emotion that is related to withdrawal from
social interactions, including unfollowing Twitter followees or engaging less in following new
people.
The results on the negative feelings also indicate that expressing loneliness or sadness
might be less attractive for followers. This is not only the case on Facebook, where there is a
strong positivity norm (Utz, 2015b) but also on Twitter (Gruzd, 2013), where positive emotions
were re-tweeted three times higher than negative emotions, suggesting people do not want to
read negative updates. These findings are also in line with prior studies on emotional contagion
on social media (Kramer, Guillory, & Hancock, 2014). The networks of people retweeting tweets
about sadness were also smaller, indicating that tweeting about sadness is less popular than
tweeting about positive things. This pattern might also indicate that people tweeting about
sadness and with smaller networks flock together on Twitter. In line with this explanation, the
effects were especially pronounced for the number of followees. It would be easy on an
asymmetric network such as Twitter to follow many people. Smaller followees networks thus
indicate that the people expressing loneliness or sadness might be less interested in expanding
their networks and/or be less active on Twitter in general. Although this is a meaningful
interpretation of our results, we would like to point out that we used the number of followers and
followees as dependent variable in a repeated measurement analysis mainly because this analysis
allows us to detect in the same step significant differences between the number of followers and
followees. This analysis should however not be misinterpreted as indicator that expressing
feelings causally affects network size. It is also possible that the network size influences the
feelings or that both processes occur simultaneously.
Although it was not the main focus of the study, but rather served as a control group, we
found some interesting patterns in the retweets: The networks of people who retweeted tweets
about loneliness were larger than the ones of people who tweeted about loneliness. In a similar
vein, people who retweeted happy tweets had larger networks (both followers and followees)
than people retweeting sad tweets. The latter might indicateagain that positive tweets are more
attractive for potential followers, but the finding on the popularity of people retweeting
loneliness tweets is somewhat puzzling. Maybe these people are very agreeable and helpful and
therefore popular. They might do it for an altruistic motive to help people with exposing their
messages to more people. These interpretations would be in line with boyd’s et al. finding that
people retweet tweets as a signal of attention/friendship or to expose less popular users to a
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
11
larger audience. That said, the reasons for retweeting (highlighted above) and the complexity of
this practice (boyd, Golder & Lotan, 2010) make retweeting by influential users deserving of
attention in a future study.
We would also like to note some limitations and strengths of the present study. A limitation
is that we collected 11,880 tweets posted in English. This means that the findings of this study
apply only to tweets written in English. Moreover, we did not take into account the influence of
culture on the expression of emotions. The people in our sample could be from different English
speaking countries or might even be non-native English speakers. There is no doubt that
behaviours regarding feelings and social interaction may vary significantly across cultures. We
recommend that the influence of culture on the expression of emotions be taken into account in
future research.
In addition, we only tracked single tweets expressing a certain feeling, but did not analyze
the subsequent and/or prior tweets of the users to examine whether they are in general more
negative in their tweets than people expressing positive feelings. However, this has been done by
researchers focusing on depression expressed in tweets (De Choudhury et al., 2013; Park, Cha, &
Cha, 2012) and was therefore not the main focus of the present paper. Based on these prior
results, we can assume that at least the clinically depressed subset of our sample tweets were
more negative in general.
Another limitation is that we only tracked the number of followers and followees, but did
not examine the network structure. Follow-up research could test whether the networks of people
expressing loneliness or sadness are not only smaller, but also less dense than the networks of
people expressing positive feelings. Due to limitations in the Twitter API, free tools do not allow
the download of complete networks anymore. The effect sizes of our findings were also rather
small, indicating that there are many other factors influencing the number of followers and
followees. However, it would also be unrealistic to expect that a single tweet results in huge
effects on network size, especially because it might not even be noticed by many followers.
A strength of the study is that we focused on two different negative feelings and that we
included two control conditions: the corresponding positive feeling and retweets of tweets
containing the respective feelings. This allows us to disentangle the effects of expressing a
genuine feeling from having an affinity to talk about negative/positive emotions.
6 Conclusion
The present study compared the Twitter networks of people expressing loneliness or sadness
with the Twitter networks of people expressing feeling loved and happy and found significant
differences in network size. People expressing negative emotions on Twitter had smaller
networks, especially less followees. These networks stayed small or became, in the case of the
people expressing sadness – even smaller over time.
References
Ackland, R. (2013). Web social science: Concepts, data and tools for social scientists in the digital age. Sage
Al-Saggaf, Y., & Nielsen, S. (2014). Self-disclosure on Facebook among female users and its relationship to
feelings of loneliness. Computers in Human Behavior, 36, 460–468. Retrieved from
http://www.sciencedirect.com/science/article/pii/S0747563214002234
Baker, L. R., & Oswald, D. L. (2010). Shyness and online social networking services. Journal of Social and
Personal Relationships, 27(7), 873–889. http://doi.org/10.1177/0265407510375261
Yeslam Al-Saggaf, Sonja Utz & Ruoyun Lin
Bazarova, N. N., & Choi, Y. H. (2014). Self-disclosure in social media: Extending the functional approach to
disclosure motivations and characteristics on social network sites. Journal of Communication, 64(4), 635–657.
http://doi.org/10.1111/jcom.12106
Bazarova, N. N., Choi, Y. H., Schwanda Sosik, V., Cosley, D., & Whitlock, J. (2015). Social Sharing of Emotions
on Facebook. In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work &
Social Computing - CSCW ’15 (pp. 154–164). New York, New York, USA: ACM Press.
http://doi.org/10.1145/2675133.2675297
Bollen, J. (2015). Modeling critical transitions in mental health from longitudinal social media data. In 2nd GESIS
Computational Social Science Winter Symposium. Cologne, Germany.
Chelune, G. J., Sultan, F. E., & Williams, C. L. (1980). Loneliness, self-disclosure, and interpersonal effectiveness.
Journal of Counseling Psychology, 27(5), 462–468.
Davis, M. H., & Franzoi, S. L. (1986). Adolescent loneliness, self-disclosure, and private self-consciousness: A
longitudinal investigation. Journal of Personality and Social Psychology. US: American Psychological
Association. http://doi.org/10.1037/0022-3514.51.3.595
De Choudhury, M., Counts, S., & Horvitz, E. (2013). Social media as a measurement tool of depression in
populations. In Proceedings of the 5th Annual ACM Web Science Conference (pp. 47–56). ACM.
Forest, A. L., & Wood, J. V. (2012). When social networking is not working: individuals with low self-esteem
recognize but do not reap the benefits of self-disclosure on Facebook. Psychological Science, 23(3), 295–302.
http://doi.org/10.1177/0956797611429709
Golden, J., Conroy, R. M., Bruce, I., Denihan, A., Greene, E., Kirby, M., & Lawlor, B. A. (2009). Loneliness, social
support networks, mood and wellbeing in community-dwelling elderly. International Journal of Geriatric
Psychiatry, 24(7), 694–700. http://doi.org/10.1002/gps.2181
Grieve, R., Indian, M., Witteveen, K., Anne Tolan, G., & Marrington, J. (2013). Face-to-face or Facebook: Can
social connectedness be derived online? Computers in Human Behavior, 29(3), 604–609.
http://doi.org/10.1016/j.chb.2012.11.017
Gross, J., Richards, J., & John, O. (2006). Emotion regulation in everyday life. … : Pathways to Dysfunction ….
Retrieved from http://media.rickhanson.net/Papers/EmotRegDaily Life.pdf
große Deters, F., & Mehl, M. R. (2013). Does posting Facebook status updates increase or decrease loneliness? An
online social networking experiment. Social Psychological and Personality Science, 4(5), 579–586.
http://doi.org/10.1177/1948550612469233
Joiner Jr, T. E. (1997). Shyness and low social support as interactive diatheses, with loneliness as mediator: Testing
an interpersonal-personality view of vulnerability to depressive symptoms. Journal of Abnormal Psychology,
106(3), 386.
Kivran-Swaine, F., Ting, J., Brubaker, J. R., Teodoro, R., & Naaman, M. (2014). Understanding Loneliness in
Social Awareness Streams: Expressions and Responses. Proc. ICWSM.
Kramer, A. D. I., Guillory, J. E., & Hancock, J. T. (2014). Experimental evidence of massive-scale emotional
contagion through social networks. Proceedings of the National Academy of Sciences, 111(24), 8788–8790.
Krause, N. (1987). Chronic financial strain, social support, and depressive symptoms among older adults.
Psychology and Aging, 2(2), 185.
Kross, E., Verduyn, P., Demiralp, E., Park, J., Lee, D. S., Lin, N., … Ybarra, O. (2013). Facebook use predicts
declines in subjective well-being in young adults. PLoS ONE, 8(8), e69841.
http://doi.org/10.1371/journal.pone.0069841
Lee, K.-T., Noh, M.-J., & Koo, D.-M. (2013). Lonely people are no longer lonely on social networking sites: the
mediating role of self-disclosure and social support. Cyberpsychology, Behavior and Social Networking,
16(6), 413–8. http://doi.org/10.1089/cyber.2012.0553
Lin, H., & Qiu, L. (2012). Sharing emotion on Facebook. In Proceedings of the 2012 ACM annual conference
extended abstracts on Human Factors in Computing Systems Extended Abstracts - CHI EA ’12 (p. 2573). New
York, New York, USA: ACM Press. http://doi.org/10.1145/2212776.2223838
Park, M., Cha, C., & Cha, M. (2012). Depressive moods of users portrayed in twitter. In Proceedings of the ACM
SIGKDD Workshop on Healthcare Informatics (HI-KDD) (pp. 1–8).
Reinecke, L., & Trepte, S. (2014). Authenticity and well-being on social network sites: A two-wave longitudinal
study on the effects of online authenticity and the positivity bias in SNS communication. Computers in Human
Behavior, 30, 95–102. http://doi.org/10.1016/j.chb.2013.07.030
Rubinstein, C., Shaver, P., & Peplau, L. A. (1979). Loneliness. Human Nature, 2(2), 58–65.
Stokes, J. P. (1985). The relation of social network and individual difference variables to loneliness. Journal of
Personality and Social Psychology, 48(4), 981–990. http://doi.org/10.1037/0022-3514.48.4.981
EXPRESSING FEELINGS ON TWITTER AND NETWORK SIZE
13
Turkle, S. (2012). Alone together: Why we expect more from technology and less from each other. New York: Basic
Books.
Utz, S. (2015a). Is LinkedIn making you more successful? The informational benefits derived from public social
media. New Media & Society . http://doi.org/10.1177/1461444815604143
Utz, S. (2015b). The function of self-disclosure on social network sites: Not only intimate, but also positive and
entertaining self-disclosures increase the feeling of connection. Computers in Human Behavior, 45, 1–10.
http://doi.org/10.1016/j.chb.2014.11.076
