1. No category

Unfriending on Facebook: Friend Request and
Online/Offline Behavior Analysis
Christopher Sibona
Second Year Research Paper directed by Associate Professor Steven Walczak
Comprehensive Exam
Information Systems
The Business School
University of Colorado Denver
christopher.sibona@email.ucdenver.edu
Contents
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
2
Literature Review and Background . . . . . . . . . . . . . . . . . . . . . . . . . .
2
2.1
Social Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
2.2
Friendship Formation & Dissolution . . . . . . . . . . . . . . . . . . . . . . . . .
3
2.3
Computer-mediated communications . . . . . . . . . . . . . . . . . . . . . . . . .
5
2.4
Online Etiquette/Netiquette . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
3
Initial Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
4
Instrument Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5
Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6
Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
7
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7.1
Friend Request . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7.2
Individual Factor Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.3
Factor Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.4
Construct Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.5
Reliability Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.6
A Comparison of Online and Offline Constructs and the Decision to Unfriend . . . 24
7.7
Final Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.8
Generated Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
9
Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
10
Implications and Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . 35
10.1 Implications for Individual Users and Business Users . . . . . . . . . . . . . . . . 35
10.2 Implications for Facebook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
10.3 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1 Introduction
1
Abstract
Objectives: Determine the role of the friend request in unfriending decisions on the social network
site Facebook.com. Find factors in unfriending decisions and find differences in the perception of
online and offline behaviors that vary depending on the unfriending decision.
Method: Survey research conducted online. 1,137 surveys about unfriending were analyzed using
exploratory statistical techniques. Survey respondents were recruited from Twitter.
Results: The research results show that the initiator of the friend request has more than their
expected share of unfriends compared to those who receive the friend request. There are online
and offline factors for unfriending decisions; the research identified six constructs to evaluate
unfriending decisions. There are 4 constructs for online behaviors (unimportant/frequent posts,
polarizing posts, inappropriate posts and everyday life posts) and 2 offline constructs (disliked
behavior and changes in the relationship). Survey respondents who said they unfriended for online
reasons were more likely to agree that the person posted too frequently about unimportant topics,
polarizing topics, inappropriate topics and everyday life topics compared to those who unfriended
for offline reasons.
1
Introduction
Facebook has over 500 million active users worldwide1 and is the most popular website in the
U.S. Facebook recently overtook Google as the most visited website on March 15, 2010 according
to webtracker Hitwise (Nuttal and Gelles, 2010). Research in social behavior indicates that the
Internet is used to maintain existing relationships, form romantic connections, and create new
online friendships (Wang et al., 2010). These online friendships are fluid; friendships are created
and dissolved on social network sites.
The word unfriend was named the word of the year by the New Oxford American Dictionary for
2009 (Goldsmith, 2009). The dictionary defined unfriend as follows: “unfriend – verb – To remove
someone as a ‘friend’ on a social networking site such as Facebook.”2 Real-world friendship
1
2
http://www.facebook.com/press/info.php?statistics
http://blog.oup.com/2009/11/unfriend/
2 Literature Review and Background
2
dissolution has been studied in a variety of contexts such as romantic relationships, marriage and
divorce (Levinger, 195), interracial friendships (Hallinan and Williams, 1987), high school and
college students (Owens, 2003). Unfriending on social networks is different in a number of ways
than real-world friendships. Facebook has a definite marker when the friendship is dissolved by
one member through the unfriend function. On a social network site one member can decide
to end the relationship and will publicly remove support by unfriending that person. Friendship
dissolution for online friendships (a.k.a. unfriending) is under-researched.
There are two major research questions that this study addresses.
1. What is the role of the friend request in unfriending decisions.
2. Can factors in unfriending decisions be found and do differences in the perception of online
and offline behaviors vary depending on the unfriending decision.
2
2.1
Literature Review and Background
Social Networks
boyd and Ellison (2007) defined social network sites based on three system capabilities. The systems: “allow individuals to (1) construct a public or semi-public profile within a bounded system,
(2) articulate a list of other users with whom they share a connection, and (3) view and traverse
their list of connections and those made by others within the system” (boyd and Ellison, 2007, p.
1). After users join a site they are asked to identify others in the network with whom they have an
existing relationship. The links that are generated between individuals become visible to others in
the local subspace. There are general social network sites (Facebook) and others that are content
focused. Social network site MySpace has a large musical component and links musicians and
musical groups to their fans, LinkedIn is a network of professional contacts and YouTube connects
people who are interested in video content (boyd and Ellison, 2007). Facebook, which is focused
on real friends (Ellison et al., 2007), is now believed to be the largest social network site (Raphael,
2 Literature Review and Background
3
2009).
Research on social network sites and on Facebook in particular cover a variety areas. Research interests include identity management (Hewitt and Forte, 2006), trust (Dwyer et al., 2007),
self-presentation (Stutzman, 2006), surveillance and privacy concerns (Gross and Acquisti, 2005),
social capital (Ellison et al., 2007). Much of the academic research on Facebook has focused on
identity presentation and privacy concerns (Ellison et al., 2007).
2.2
Friendship Formation & Dissolution
Friendships are formed and maintained because they are rewarding to individuals (Wright, 1984).
Friendship models have been developed to describe how friendships are created (Hallinan, 1979).
Friendships tend to be formed by people who share certain similarities (such as values) (Lea and
Duck, 1982; McPherson et al., 2001). People tend to create friendships with those who share a
similar race and ethnicity followed by age, religion, education, occupation and gender and roughly
in that order (McPherson et al., 2001). The largest portion of friendships that are formed with
those who are not family members are through organizational structures (McPherson et al., 2001).
Schools, work, and geographic location are major factors in how relationships are formed. Hallinan
describes the process of friendship formation as “a sequential process having four elements. First,
P must desire to have O as a friend (attraction). Second, P must initiate a move to establish a
friendship with O. Third, O must recognize P’s overture of friendship. Fourth, O must reciprocate
P’s offer of friendship” (Hallinan, 1979, p. 194). The initiator of the friendship request tends to
have lower status than the recipient (Hallinan, 1979).
Friendship formation in the real world has more nuance than in the online world. The initiator
of the friend request may communicate the desire to be friends with varying degrees of directness
(Hallinan, 1979). Those who initiate the friendship in less direct ways can avoid embarrassment
and rejection should the request not be accepted. The online world (on a site like Facebook) lacks
this nuance and makes it very clear that one person requests the other’s friendship through the
2 Literature Review and Background
4
visible friend request.
On Facebook one person initiates a “friend” request and another person receives the request.
The request for friendship is easily identified in the online world because there is a marker for that
request. The receiver can choose to accept the friendship request or choose to ignore the request.
If the receiver accepts the request then the two become “friends” on Facebook. Social network
sites use different terms for the relationship; Facebook uses the term “friends” but others use the
term “contacts,” “fans,” or “follower” (boyd and Ellison, 2007). Visible virtual links are generated
between the dyad’s online identities. The online friendship process mirrors friendship formation
in the real world with the addition of markers like the visible request and visible link connecting
the dyad.
Friendship dissolution is not the same process of friendship formation in reverse and is distinctly
different (Duck, 1982). Some friendships end in conflict but most simply fade away (Sprecher and
Fehr, 1998). Friendships do not require the other person’s permission to end the relationship in
either the online or offline world (Baxter, 1979). You need permission to be someone’s friend
on Facebook; however, no permission is needed to end the relationship. One person can simply
choose to “unfriend” the other person. In most cases the person who was unfriended does not
receive notification that they have been unfriended. There are applications available on Facebook,
such as “unfriend finder,” that notify users when they are unfriended.3
There are both internal and external barriers to friendship dissolution (Bushman and HoltLundstad, 2009). External barriers are those that originate outside the individual and make the person feel as though the relationship must be maintained. External barriers include social (church,
family), financial ties, and physical proximity (coworkers, neighbors, etc.). Internal barriers are
endogenous and drive the person to maintain the relationship and include religious beliefs, selfidentity (sociability) or a personal sense of commitment.
Online social networks, with their visible links between members, may make it difficult to
end a relationship. Unlike real world relationships that may simply fade without either member
3
http://www.facebook.com/unfriendfinder
2 Literature Review and Background
5
making a conscious decision about the dissolution, online unfriending is a conscious and public
decision. Researcher Steve Duck calls the public declaration of the end of a relationship in the
real world “grave dressing” (Duck, 1982). In Facebook the unfriending can be a signal to others in
the network that the particular relationship is over because the links between members are visible.
Others in the network who share links with the dyad can notice that a link previously present has
been removed.
2.3
Computer-mediated communications
Computer-mediated communications have been researched to determine how reduced cues impact
online interactions (Haythornwaite, 2002). Online posts provide different cues than face-to-face
conversations. Walther (1996) examined seemingly contradictory views of online communication
when he classified communication as impersonal, interpersonal or hyperpersonal. Computer operators, who were early users of online communication systems, initially sent simple messages
to each other to coordinate tasks. When these messages were examined, researchers found that
the messages reduced interpersonal affect and group solidarity and were more task oriented than
face-to-face meetings (Walther, 1996). Researchers noticed that as users of computer-mediated
communication systems used these systems over time and in different contexts that the communication grew more interpersonal. The messages contained less social information than face-to-face
meetings, as they lacked nonverbal cues, but over time the users generated additional methods to
help decipher the interpersonal cues. As users had increased expectations of future online interactions through computer messages they attempted to learn more about the person with whom they
were communicating. These communications tended to be more interpersonal than the computer
operator messages because they were expected to continue in the future vs. the single task-oriented
messages.
As computer usage has evolved to include more recreational interactions (games, chat, etc.)
the level of affect can exceed levels found in face-to-face interactions and may be classified as
2 Literature Review and Background
6
hyperpersonal (Walther, 1996). Users of these systems can develop stereotypical impressions
of their online partners because certain social information is not present. Receivers of online
messages will fill in the missing details themselves by creating a more complete but distorted
version of the sender. Online communications users can selectively present the information that
they wish to reveal (Walther, 1992; Riva, 2001). User are allowed more time in which to generate
a response to a message and thus may respond more deliberately. The receiver of these messages
can create such complete and idealized images of the person with whom they communicate that
when they meet in real life they can feel disappointment that the realized person did not meet their
imagined expectations (Grimes, 1992).
Haythornwaite (2002) ties these three communications mechanisms by examining the strength
of the tie between the dyad. When the ties are weak, as in the computer operators sending messages
about the systems, then the dyad tends to be more dependent on organizationally established means
of communication. When the ties are strong communicators will influence each other and adapt
the system for additional uses. Strong communicators can expand their use of the systems beyond
the original context or work towards its dissolution if they feel it impedes their communication.
When ties between the dyad are weak and new forms of communication are introduced then the
dyad is susceptible to dissolution. Weakly tied communicators tend to be more passive in their use,
less likely to resist, and more accepting of mandated technologies.
People tend to have complex views about new communication technologies as they are adopted;
some are skeptical of the technology while others embrace it (Baym, 2010). Some believe that the
new technology devalues face-to-face communication and express concern that communication
will grow increasingly shallow. Others will embrace the technology and tout its benefits to create
stronger relationships and new opportunities to interact with new people. As the technology is
adopted by greater numbers, people see the the pros and cons of new media and accept it for what
it provides. Baym (2010) discusses how concerns about the adoption of new technologies have
persisted through centuries; technologies such as the telegraph, telephone, email, instant messaging, mobile phones and today’s Internet have raised concern among users and have been accepted
2 Literature Review and Background
7
by society to a large degree. New communication technologies, such as online social networks, are
expected to be influenced by existing communication mechanisms and be adapted to fit the specific
context as users gain experience with the technology over time (Yates and Orlikowski, 1992).
Most online interactions tend to be between people who have met face-to-face or talked on the
telephone and not with random “strangers” who have met online (Baym et al., 2004). Relationships
can be maintained through multiple media channels and generally do not require only one channel
for communication. Baym et al. (2004) also argues that there is no “inherent” trade-off between
more traditional “high-quality” communication mechanisms such as face-to-face conversation and
interactions on the Internet that may be considered more “superficial.” The Internet is not a distinctly separate space from offline interactions where individuals in one space can not interact with
those in the other space. The reality is that people no longer discuss how the telephone has negatively impacted real-life relationships and lessened the value of face-to-face communication. The
telephone is embraced as a communication tool for which certain messages are appropriate. There
are those who believe there is a danger that online communication will hurt real-life relationships
(Bernstein, 2009). If Baym’s communication channels are extended to include social network sites
such as Facebook then it is likely that as adoption increases, becomes pervasive and routinized that
social network site users will be able to appropriately determine which messages are best suited
for these social networks. Social network sites will be valued as an appropriate channel for some
messages while other channels will be valued for other messages.
2.4
Online Etiquette/Netiquette
There are often rules that govern what can be posted in a given Internet-based forum. These rules
can be both formal and informal. Formal rules are written and explicitly identify speech that is considered acceptable and unacceptable. Informal rules place normative barriers to certain behaviors;
approximately 15% of all messages in Usenet forums are considered conduct-correcting (Smith
et al., 1997). Forums can have specific rules about acceptable posting behavior and can control
2 Literature Review and Background
8
the content of the forum through the use of moderating practices (Chua, 2009). Designated forum
moderators can screen, reject and remove messages that violate a forum’s posting policy. There
are trade-offs between the use of moderators to control the types of messages that can be posted
to forums. Moderators can increase the percentage of on-topic posts compared to unmoderated
forums by screening posts that are off-topic or otherwise problematic (Chua, 2009). Moderated
forums can increase the sense of collective identity by keeping a community on-topic. Moderated
forums can also reduce the free interchange of ideas and silence speech in forums where multiple
groups with separate identities may wish to discuss a topic of overlapping interest (Chua, 2009).
There can be trade-offs between a stronger community identity and the free exchange of ideas
and forums will value these trade-offs differently. Moving a forum from an unmoderated policy
to a moderated policy is considered a final step after normative measures fail to curtail unwanted
behavior (Blanchard, 2004).
Facebook has formal rules which govern posting behavior4 and allows users to act as the moderator for their personal Wall. The rules include prohibitions against certain commercial speech
(spam and multi-level marketing), content, illegal activity, and bullying behavior. Prohibited content includes that which is: “hateful, threatening, or pornographic; incites violence; or contains
nudity or graphic or gratuitous violence.” The legal prohibitions include speech that is: “unlawful,
misleading, malicious, or discriminatory.” Behavior that is directed toward an individual which
is used to: “bully, intimidate, or harass any user” is prohibited. Facebook says that, “The Wall is
the center of your profile for adding new things, like photos, videos, notes and other application
content.”5 Facebook users can can act as the moderator to the messages that others post on their
personal wall.6 Users can restrict whether others can post on their wall through privacy settings
and remove offending posts. Facebook currently does not provide the ability to review posts before
they are posted to the Wall. Facebook provides similar tools that are available to forum moderators;
Facebook users can remove offensive posts from their wall and remove the individual completely
4
5
6
http://www.facebook.com/terms.php
http://www.facebook.com/help/?faq=13153
http://www.facebook.com/help/?page=443
3 Initial Model
9
from the local subspace by choosing to unfriend the person.
3
Initial Model
An initial model was developed to examine online and offline reasons for unfriending (Figure 1
on page 12). The questions were developed from a variety of sources. Literature review included
a variety of communications based research articles about conversation, friendship and dissolution (Oliver, 1961; Owens, 2003; Bushman and Holt-Lundstad, 2009; Parks and Floyd, 1996;
McPherson et al., 2001; Hartup and Stevens, 1999; Riva, 2001; Sternberg, 2000). The popular
press has written numerous articles about unfriending e.g.: Boston Globe (2009) “To ’Unfriend’
or Not To ’Unfriend,”’7 Bloomington Pantagraph (2009): “In the world of Facebook, it’s OK to
unfriend some people,”8 Cosmopolitan (undated) “10 Signs You Should Unfriend Someone on
Facebook.”9 Blogs also offer advice on who to unfriend e.g.: From Mashable: “12 Great Tales of
De-friending,”10 from Gawker: “The Eight Types of People to unfollow on Twitter or Defriend on
Facebook,”11 from The Frisky: “8 Signs You Should Unfriend Someone on Facebook,”12 from associatedcontent: “When should you ’Unfriend’ Someone on Facebook”13 , Timesunion.com (2010)
“7 reasons to unfriend someone on Facebook.”14
The survey questions were designed through literature review and informal interviews with a
variety of Facebook users to determine the types of friends who they unfriended and online posting behavior. Open ended questions were asked of 10 Facebook users to determine what kinds of
7
http://www.boston.com/jobs/news/jobdoc/2009/12/to_unfriend_or_not_to_
unfriend.html
8
http://www.pantagraph.com/news/article_119e3685-405e-5eb4-859f-71b3d0c1e58e.
html
9
http://www.cosmopolitan.com/advice/tips/signs-you-should-unfriend-someone
10
http://mashable.com/2008/11/25/social-network-defriending/
11
http://gawker.com/5500413/the-eight-types-of-people-to-unfollow-on-twitter
-or-defriend-on-facebook
12
http://www.thefrisky.com/post/246-signs-you-should-unfriend-someone-on
-facebook/
13
http://www.associatedcontent.com/article/5663516/when_should_you_unfriend_
someone_on.html
14
http://blog.timesunion.com/kristi/29665/7-reasons-to-unfriend-someone-onfacebook/
4 Instrument Design
10
people they have friended and unfriended, what kinds of posts they see on Facebook, how they
were affected by an unfriending, have they unfriended someone on Facebook, and, if so, why. The
researchers also asked what kinds of posts for which they may have unfriended. This stage of
question development focused solely on online reasons for unfriending. A question was posted
on an internet forum (beginner triathlete) that asked for reasons why forum users believed they
were unfriended and forum users unfriended others. 56 responses were received from the forum
to validate online and offline reasons for unfriending. After an initial survey was developed, five
people were asked questions orally by the researchers to determine appropriateness and completeness. The online survey was pretested by six Facebook users to determine whether there would be
difficulty completing the survey, survey length, completeness, etc.
Twenty one questions were developed to examine online reasons for unfriending and twelve
questions for offline reasons - see Table 1. and Table 2. . The questions are grouped into subgroups
based on relational factors and interpretation. The online questions are grouped into three groups:
posting frequency, inflammatory topics and banal topics. The offline questions are grouped into
two subgroups: disliked behavior/personality and changes in the relationship to reflect offline reasons for unfriending. Exploratory factor analysis will be used in the study to determine appropriate
constructs and a final model will be developed.
4
Instrument Design
The survey was conducted solely on the Internet using a commercially available survey tool. The
survey questions are a combination of established questions from previous studies and new questions to examine friendship dissolution in online settings plus demographic questions. The survey
opens with a cover letter to introduce the survey using the Total Design Method by Dillman (1978).
4 Instrument Design
11
Table 1. : Online Questions and Sources
Question
Frequency
Politics
Religion
Sex
Money
Unflattering
Racist
Sexist
Swearing
Inappropriately
Unimportant
Eating Habits
Exercise Habits
Sports Scores
Celebrities
Purchases
Promotion
Job
Child
Spouse
Pets
Source
Interview, Oliver, 1961, Boston Globe, Gawker
Interview, Oliver, 1961; McPherson et al., 2001; Chua, 2009; Gilbert and
Karahalios, 2009, Mashable, The Frisky
Interview, Oliver (1961); McPherson et al. (2001); Chua (2009), Gawker,
The Frisky
Mashable, Times Union, Gawker
Owens (2003)
Interview, Oliver (1961); Chua (2009), Mashable, associated content
Interview, Chua (2009); Nielsen (2002), Facebook Policy on hate speech
Nielsen (2002)
Interview, Oliver, 1961; Riva, 2001
Interview, Oliver (1961); Riva (2001), associated content, Gawker, The
Frisky
Interview, Oliver (1961); Chua (2009), Cosmopolitan, Gawker
Interview, Randaza, 2009, Cosmopolitan, Times Union
Randaza, 2009, Cosmopolitan, Times Union
Interview, Mashable
Randaza, 2009
Randaza, 2009
Interview, Riva (2001); Randaza (2009), Mashable, Times Union, The
Frisky
Oliver (1961), Times Union, Mashable
Interview, Cosmopolitan
Interview
Interview
4 Instrument Design
12
Fig. 1: Initial Model
4 Instrument Design
13
Table 2. : Offline Questions and Sources
Question
Did Misdeed
Dislike
Behavior
Betray
Broke Rule
Personality
Trust
New Information
Incompatible
Friends
Geography
Divorce
Romantic End
Source
Metts (1994)
Owens (2003); Sprecher and Fehr (1998)
Owens (2003); McPherson et al. (2001); Metts (1994)
Sprecher and Fehr (1998); Metts (1994)
Metts (1994)
Owens (2003)
Owens (2003); Metts (1994), associated content
Sprecher and Fehr (1998); Metts (1994)
Owens (2003); Sprecher and Fehr (1998), associated content
Owens (2003); Sprecher and Fehr (1998); McPherson et al. (2001)
Sprecher and Fehr (1998); Metts (1994)
Sprecher and Fehr (1998); Metts (1994)
The survey screens users to determine if the person is over 18 and a Facebook user. The survey
was approved by the Colorado Multiple Institutional Review Board at the University of Colorado
Denver (protocol #10-0369).
The survey is divided into four sections. Part one of the survey asks whether the survey respondent has unfriended someone on Facebook, and, if so, asks questions about that unfriending.
Part two of the survey asks whether the survey respondent has ever been unfriended on Facebook,
and, if so, asks questions about that unfriending. Part three of the survey asks questions about
Facebook usage, and part four asks demographic questions. The behavioral sections used 7-point
Likert-type questions and were presented in a randomized order. Demographic questions and categorical questions were multiple choice. The survey was designed to take approximately fifteen
minutes to complete. The path through the survey depended on answers to the questions. For
example, if the respondent said they have never unfriended a person the survey tool would advance
to the next section.
Part one of the survey asked questions about the type of person unfriended, whether it was for
online or offline behavior, questions about the friendship and questions about online and offline
behavior. Part two mirrors part one of the survey and asks questions about the type of person who
5 Data Collection
14
unfriended the survey respondent, their perception of whether it was for online or offline behavior,
questions about the friendship and questions about their offline behavior. Part two adds additional
questions to part one to determine how the survey respondent was affected by the unfriending.
Part three asks questions about how many friends the survey respondent has, how many people
they have unfriended, how many people they regularly interact with, and questions about their
online posting behavior. Part three also asks questions about satisfaction, perceived usefulness and
perceived ease of use of Facebook. Part four asks demographic questions: age, gender, education,
the number of years of social network use and whether the person lives in the United States of
America.
The questions about online and offline behavior are largely exploratory. Components were not
identified during the survey design and factor analysis would later partition the data into meaningful subgroups.
5
Data Collection
Survey respondents were recruited from Twitter. Twitter respondents were gathered by screening tweets that had the term “unfriend,” “defriend,” or “unfriending.” Tweet messages, known
as tweets, that met a screening criteria were sent replies inviting the person to take the survey
about unfriending. Tweets that were not considered for the study include: non-English tweets,
unfriending tweets that were not about people but about something else (TV shows, politicians,
states, etc.), tweets about news articles about unfriending, tweets that appeared to be jokes (some
that specifically included LOL), inflammatory tweets, and some short tweets that just included the
word “unfriend.” Tweets were only sent to people who were writing about unfriending in English.
That is, there are tweets that were written in a different language (e.g. German) with the English
word “unfriend” in it and that person would not receive a reply asking them to take the survey.
Occasionally the tweet reply was retweeted by those who received the initial tweet. The sample
did not seek out expert opinions on social network sites because those responses might be biased.
6 Methodology
15
There is not a random sample in this research; a convenience sampling method was used to recruit
participants.
Surveys were collected between April 17th and July 17, 2010. A total of 2,084 surveys were
started and 1,137 were completed. A total of 4,961 recruitment tweets were sent to Twitter users
during the recruitment period. 54.6% of those who started the survey completed the survey and
the mortality rate was 45.4%. The response rate for those who completed the survey is 22.9%
(1,137/4,961). The overall response rate, that is people who came to the survey and started the
survey, is 42.0% (2,084/4,961). The total number of tweets sent from the recruitment account is
6,935. The tweets that were sent beyond the 4,961 tweets were responses to questions about the
survey, responses to thanks, questions about whether this account was a bot, etc. 71.5% of the
the tweets were sent were the recruitment tweet. The Twitter account created, UnfriendStudy, was
only used for the purpose of this study. The analysis uses all the data collected and is not limited
to completed surveys only.
6
Methodology
The raw data was collected from a commercially available survey tool (Survey Monkey) and analyzed with SPSS version 18. The survey is largely exploratory and used methods such as factor
analysis to find commonalities among the questions asked. Factor analysis was used to partition
questions into meaningful groups. Constructs were then generated based on the factor analysis
and interpretation of the results and Cronbach’s alpha measure of reliability was calculated. Crosstabulation techniques was used to determine the differences in actual frequencies vs. expected
frequencies between different groups (for example those who unfriend for online vs. offline behavior) and determine the role of the friend request in unfriending decisions (the first research
question). MANCOVA was used to determine whether survey takers had higher levels of agreement about posting topics or offline behavior based on their reason to unfriend (online vs. offline)
which is the second research question.
6 Methodology
16
The constructs unimportant/frequent posts, polarizing posts, inappropriate posts, and everyday
life were generated and used to examine online posting behavior. The constructs disliked behavior
and change were generated and used to examine offline behavior. MANCOVA was used to determine the difference in the online and offline constructs based on the person’s unfriending decision
(online vs offline).
Statistical tool selection is based on the appropriateness to the model and unit of analysis.
MANOVA is used to analyze multiple dependent variables that are correlated with each other
in a low to moderate level (Leech et al., 2008). Cross-tabulation techniques are used to calculate
the expected frequency in a cell and compare the expected cell count to the actual cell count found
in the data (Joseph F. Hair et al., 2006). MANCOVA is used to adjust for difference between the
groups based on another typically interval-level variable called the covariate (Leech et al., 2008).
The analysis uses unfriending decision (online vs. offline) as an independent variable in this
analysis. The data in this analysis is about unfriendings that have occurred and does not try to
generate a probability function for a future potential unfriending. That is, this research does not
attempt to define a probability function to determine whether the dyad will remain friends in the
future; this research only captures unfriendings that have already occurred. 84.0% of survey respondents could identify whether they unfriended a person for online or offline reasons with the
rest unsure. The analysis presumes that the unfriending decision reflects why someone choose to
unfriend the person. If differences are seen between the constructs for the online and offline behaviors (dependent variables) and the decision to unfriend (independent variable) then it is likely that
those constructs are relevant in actual unfriending decisions and provide a basis of comparison.
The results are most informative when the survey respondent said they unfriended someone for
online reasons and differences are seen in the online posting behaviors compared to those who
selected offline reasons or in factors where no differences are found. The decision to unfriend is
an independent variable in the MANCOVA analysis.
Exploratory methods and interpretation of the results can lead to subsequent hypothesis generation that can be confirmed or disproved by additional data collection (Joreskog, 1969). If con-
7 Results
17
firmatory methods are to be employed on a single data set then the dataset should be randomly
divided into two sets: one set to generate the hypotheses and the other to confirm or disprove
the hypotheses (Joreskog, 1969). Confirmatory research starts with a set of a priori hypotheses
about a topic, develops a research design to test the hypotheses, data collection and analysis of the
data (Jaeger and Halliday, 1998). Exploratory methods can be used to generate hypotheses which
can be subsequently tested using confirmatory methods (Jaeger and Halliday, 1998). The goal of
exploratory research is to gain new insights into a phenomenon and develop testable hypotheses.
7
7.1
Results
Friend Request
Friend requests were analyzed to determine if there were differences in unfriending behavior based
on who initiated the friend request. A total of 1553 friend requests were analyzed based on the
survey respondent’s decision to unfriend someone and the survey respondent’s own unfriending.
The survey respondent was asked to indicate who initiated the friend request in both their own
decision to unfriend a person and an unfriending that happened to them. The survey included three
choices, “I asked this person to be my friend,” “This person asked me to be their friend,” and
“I don’t remember who asked.” The results show that statistically significant differences in the
expected and actual cell counts exist. The Chi-square test was 55.9 with 5 degrees of freedom; the
two-tailed significance is .0001 which indicates that there are differences in unfriending behavior
based on friend requests. The results are show in Table 3. There are three findings:
7 Results
18
1. Survey respondents were more certain of who initiated the request when they did the unfriending compared to survey respondents who were being unfriended. The don’t knows in
the table are greater than expected when the survey respondent was unfriended by someone
compared to when the survey respondent chose to unfriend someone.
2. Survey respondents who initiate the friend request are unfriended more often than expected.
The number of unfriended by when the survey respondent initiates is greater than the expected count.
3. Survey respondents who receive the friend request make more than the expected number of
unfriend decisions. The number of unfriend decisions when the other person initiated the
friend request is greater than the expected count.
Table 3. : Friend Request
I initiated
Other
initiated
DK
Total
Unfriend Decision
156
593
318
1067
Expected
203
535
329
Unfriended By
140
185
161
Expected
93
243
150
486
1553
7.2
Individual Factor Differences
The individual factors were analyzed to determine how many people agreed with the statement
about the person they unfriended and are shown in descending order by online or offline factors
in Table 4. The individual factors for overall agree were analyzed using frequency statistics. The
individual factors for online and offline factors were analyzed using cross-tabulations on the questions to determine if there are differences depending on whether an online or offline reason was
7 Results
19
chosen by the respondent. The significance level shown is whether the difference in response between those who choose online or offline reasons was statistically significant for that question. The
number of respondents (N) is shown for the cross-tabulation analysis. For cross-tabulation to be
reliable the number of expected responses in a cell should be five or greater (Joseph F. Hair et al.,
2006). All cells met the minimum expected cell count (5) to be considered acceptable. The analysis groups responses from strongly agree to agree for the agree category and from strongly disagree
to disagree for the disagree category. The groups were developed in this manner to compare survey
respondents who had a higher level of agreement with the questions.
The mean column represents the 7-point Likert scale mean for the entire dataset. That is, survey
responses for somewhat disagree, don’t know/not sure, and somewhat agree are included in the
mean statistic (unlike the comparative analysis above). The mean indicates the level of agreement
the respondents felt when they unfriended a particular person and not how they feel about the
appropriateness of any given question. That is, someone may find racist speech abhorrent in theory
but the friend they chose to unfriend did not use racist speech and they were unfriended for some
other actual reason and not a theoretical one. The higher mean indicates that survey respondents
agreed more often with that reason to unfriend compared to those with lower means.
An example of individual factor differences is provided for political posts. Overall, 26% of
survey respondents said that the person they unfriended posted too often about politics; % agree
online shows that survey respondents who said they unfriended for online reasons agreed that the
person posted too often about politics 33% of the time. Those who said they unfriended for offline
reasons said that the person posted too often about politics 11% of the time. The difference between
the level of agreement for political posts for those who indicated they unfriended for online reasons
vs. offline reasons is statistically significant at the .001 level.
7 Results
20
Table 4. : Individual Factor Differences
Question
%
Overall
Agree
Unimportant
Inappropriate
Posting Frequency
Politics
Religion
Job
Promotion
Sex
Swear
Sexist
Racist
Money
Spouse
Purchases
Celebrities
Eating
Child
Exercise
Pets
Sport Scores
65
44
34
26
20
19
18
15
14
13
12
12
11
10
9
8
8
6
5
5
Personality
Behavior
Dislike
Did misdeed
Broke rule
Trust
Betray
New Information
Incompatible
Friends
Geographic
Distance
Romantic End
Divorce
78
68
67
64
42
39
35
34
29
Mean % Disagree
online
%
Agree
online
% Disagree
offline
Online Reasons for Unfriending
4.58
28
72
52
3.63
46
54
78
3.34
58
42
83
2.91
67
33
89
2.72
75
25
88
2.64
80
20
83
2.40
77
23
91
2.40
83
17
89
2.38
84
16
91
2.36
83
17
97
2.25
84
16
96
2.35
87
13
91
2.29
89
11
91
2.28
90
10
91
2.20
90
10
95
2.24
91
9
93
2.11
91
9
94
2.09
94
6
94
2.03
95
5
94
1.96
94
6
98
Offline Reasons for Unfriending
5.06
28
72
13
4.60
43
57
15
4.56
47
53
13
4.47
56
44
6
3.51
76
24
29
3.44
82
18
23
3.32
84
16
27
3.26
72
28
54
3.13
76
24
58
%
Agree
offline
Sig
N
48
22
17
11
12
17
9
11
9
3
4
9
9
9
5
7
6
6
6
2
.001
.001
.001
.001
.001
.337
.001
.036
.018
.001
.001
.081
.478
.540
.024
.538
.254
.736
.370
.021
731
750
720
729
726
717
748
744
742
734
744
718
706
720
737
710
712
729
720
741
87
85
87
94
71
77
73
46
42
.001
.001
.001
.001
.001
.001
.001
.001
.001
716
671
674
728
675
690
664
636
617
28
3.15
75
25
65
35
.007
628
18
5
2.49
1.85
92
98
8
2
64
91
36
9
.001
.001
685
639
7 Results
7.3
21
Factor Analysis
Factor analysis was performed to generate appropriate components for the online and offline questions. Factor analysis provides a method to condense the information from a number of original variables into a smaller set with minimal losses of information (Joseph F. Hair et al., 2006).
Principal component analysis was used to determine the number of factors for online and offline
decisions based on Eigenvalues greater than 1. The factors were rotated using the Varimax function to determine factor loadings. Component groupings were then analyzed and named according
to the questions in the group. The constructs for online behavior for unfriending are: unimportant/frequent posting, polarizing posts, inappropriate posts, and everyday life posts. The constructs
for offline behavior for unfriending are: disliked behavior and changes in the relationship. Six
questions had cross loading scores above .40; incompatible friends, swearing, sexist, sex, racist,
and frequency. Table 5. shows factor loadings above a .40 threshold for the online and offline
factors.
The overall model fit was assessed and is considered acceptable. KMO measure of sampling
adequacy for the questions is .933 and is considered meritorious by Hair (2006). The six factor
loadings explain 63.5% of the variance for the factors. Factor analysis is considered acceptable for
social science research where more than 60% of the variance is explained (Joseph F. Hair et al.,
2006). Bartlett’s Test of Sphericity is statistically significant for the factor models at the .001 level.
7.4
Construct Creation
Constructs were generated based on the factor analysis results and interpretation of the factors.
Three questions were moved to a more appropriate construct. Posting racist statements was moved
to the inappropriate construct from the polarizing construct. New information about the relationship and new incompatible friends were moved to the change construct from the behavior
construct. Table 6. shows the resolution.
7 Results
22
Table 5. : Factor Analysis
Question
purchases
exercise
eating
money
pets
job
celebrities
sport scores
child
spouse
promotion
did misdeed
dislike
behavior
trust
personality
betray
broke rule
new information
incompatible
friends
swear
inappropriately
sexist
unflattering
sex
politics
religion
racist
divorce
romantic end
geographic
distance
unimportant
frequency
C1
.768
.767
.755
.726
.707
.706
.666
.657
.652
.587
.573
C2
C3
C4
.858
.826
.826
.773
.768
.759
.759
.564
.446
.404
.424
.467
C5
C6
.420
.658
.652
.570
.534
.515
.768
.690
.538 .543
.728
.691
.591
.424
.726
.636
7 Results
23
Table 6. : Construct Descriptives
Measure
Abbr
Unimportant/Frequent
Polarizing
Inappropriate
UF
Everyday Life
EL
Behavior
BH
Change
CH
PO
IN
Questions
Cronbach’sNum Mean
Alpha
of
Questions
Online Constructs
unimportant, frequent
.693
2
4.01
politics, religion
.766
inappropriate, sex,
.826
swear, sexist, racist,
unflattering
exercise, purchases,
.917
eating, money, job,
celebrities, pets, sports
scores, promotion,
child, spouse
Offline Constructs
did misdeed, dislike,
.920
behavior, personality,
trust, betray, broke rule
divorce, romantic end,
.677
incompatible friends,
change in geographic
distance, new
information
Valid
(listwise)
7.5
Std. Dev
N
2.01
1141
2
6
2.86
2.74
1.95
1.66
1064
1140
11
2.47
1.51
1125
7
4.26
1.94
1096
5
3.07
1.66
1047
969
Reliability Results
The Cronbach’s alpha for the constructs were calculated. Four of the six constructs are considered
reliable; Cronbach’s alpha measures above .70 are considered acceptable (George and Mallery,
2003). Two constructs: Unimportant/frequent posting (alpha = .693) and change (alpha = .677) are
below the .70 threshold. Table 6. show the reliability of the six constructs and number of questions
in the construct.
7 Results
7.6
24
A Comparison of Online and Offline Constructs and the
Decision to Unfriend
To compare differences in the means of the six constructs (EL, UF, PO, IN, BH and CH) based on
whether a person selected an online or offline reason for unfriending MANCOVA results were analyzed. The covariates for this procedure are the length of the friendship, how often the friend was
seen in the last year, how many friends they have in common, age, gender, education, years in social networks and whether the person lives in the US. The number of responses in the MANCOVA
analysis are: 323 for online reasons for unfriending and 195 for offline reasons. Multivariate test
results indicate that the covariates: friend length (p = .008), friend seen (p = .001), gender (p =
.002) are significant at the .05 level.
The analysis uses unfriend decision (online vs. offline) as the independent variable to understand the how online posts and offline behaviors are reflected in the decision to unfriend. Table
4.
shows how many people agree that the person posted too often about a topic overall but we do
not know if that difference is important for the constructs generated in the factor analysis. The
MANOVA results (no covariates) show that UF, PO, IN, BH and CH are statistically significant (p
< =.001); EL is the only construct that statistically significant above the .001 level. EL is significant
at the (p = .014). The MANOVA results contain no covariates so can not vary based on additional
interval-level variables. MANCOVA will determine whether a construct mean is statistically different when the person selected an online or offline reasons for unfriending. If differences are seen
in the construct means that depend on the reason for unfriending then that difference is likely to be
important in the unfriending decision. If we see no differences in the means of the constructs then
that construct is not statistically significant.
Table 7.
shows a brief description of the sample population and compares the population to
known Facebook population statistics. Age, gender, and whether the survey respondent lives in the
U.S. are among the several covariates used in the comparison analysis.
Table 4.
shows how people differ in assessing a person’s online posts depends on whether they
7 Results
25
selected an online or offline reason for unfriending. For example, 65% of people agree that the
person they unfriended posted about unimportant topics. The difference between the number of
people who agreed that the person posted too often is statistically significantly different by the unfriending decision (online vs. offline). We can consider the baseline for agreement about posting
on unimportant topics to be 48% because that is the percentage of those who said their unfriending
decision was based on offline reasons. A statistically significant difference is seen between the
offline and online reasons for unfriending; 72% of survey respondents who unfriended the person
based on the person’s online behavior said that the person posted unimportant topics. The MANCOVA analysis is comparing the differences in similar ways but with the constructs, covariates and
in a multivariate manner with multiple dependent variables depending on the independent variable:
the unfriending decision.
The results show that there are statistically significant increases in agreement that a person
posted too often about unimportant topics, polarizing topics, inappropriate topics and everyday
life topics when the survey respondent says they unfriended for online reasons compared to those
who unfriend for offline reasons. When a person indicates that they unfriended someone for offline
reasons they disliked their behavior and experienced a larger change in their relationship compared
to those who selected online reasons for unfriending.
The online constructs all had higher means when the person said they unfriended for online
reasons. This indicates that people who select online reasons for the unfriending decision agreed
more often that the person posted too often about unimportant topics, polarizing topics, inappropriate topics and everyday life topics compared to those who unfriended for offline reasons. The
offline constructs all had higher means when when the person said they unfriended for offline reasons. Survey respondents who selected offline reasons for the unfriending decision agreed more
often that they disliked the person’s behavior or experienced a larger change in the relationship
compared to those who unfriended for online reasons.
Covariates in the analysis adjust for differences between the groups; in this analysis the covariates adjust for the unfriending decision (online vs offline) and the construct under analysis. Table 8.
7 Results
26
shows the magnitude and direction of B parameter of the covariate and its effect on the construct
mean. The covariates show:
7 Results
27
Polarizing Posts
• The more often the dyad saw each other in the last year the less likely the survey respondent
would agree that the person they unfriended posted about polarizing topics compared to who
saw each other less often.
Uninteresting/Frequent Posts
• Increased length of the dyad’s friendship decreased the survey respondent’s likelihood that
they would agree that the person they unfriended posted about unimportant topics.
• Women were less likely to agree that the person they unfriended posted about unimportant
topics.
• The longer the survey respondents used social networks the more likely the survey respondent would agree that the person they unfriended posted about unimportant topics.
Disliked Behavior
• The more often the dyad saw each other in the last year increased the survey respondent’s
likelihood of citing behavior (did misdeed, dislike, etc.) as a reason for unfriending compared to who saw each other less often.
• Women were more likely to agree that they disliked the behavior of the person they unfriended.
Change in Relationship
• The longer the survey respondent used social networks the more likely the survey respondent
would agree that the person they unfriended was due to a change in the relationship.
• Older survey respondents were less likely to agree that the person they unfriended was due
to a change in the relationship.
7 Results
28
Table 7. : Frequencies
Collected by the study
Category
N
Valid %
External comparison
group
US Facebook Users %
21.0
43.0
26.5
7.5
2.0
10
29
23
18
13
7
Age
13-17
18-25
26-34
35-44
45-54
>55
M
F
Category
Yes
No
Online
Offline
Don’t Know
Online
Offline
Don’t Know
239
438
302
85
23
Gender
369
32.5
768
67.5
N
Valid %
Live in the U.S.A.
789
69.4
348
30.6
Decision to Unfriend by Behavior
889
57.0
420
26.9
250
16.0
Perception of Unfriending by Behavior
135
27.8
117
24.1
234
48.1
44.5
55.5
U.S. vs. Non-U.S.
Facebook Users
70
30
• Age Data from Facebook.com, July 2010.
Data presented by Inside
Facebook
http://www.insidefacebook.com/2010/07/06/
facebooks-june-2010-us-traffic-by-age-and-sex-users-aged-18/
-44-take-a-break-2/
• Gender Data from Facebook.com, January 2010 of facebook users 18+.
Data
presented
by
Inside
Facebook
http://www.insidefacebook.
com/2010/01/04/december-data-on-facebook%E2%80%
99s-us-growth-by-age-and-gender-beyond-100-million/
• Country data from Facebook. http://www.facebook.com/press/info.php?
statistics
7 Results
29
Table 8. : MANCOVA RESULTS
Construct
Online
mean
Offline
mean
Sig.
Mean
difference
online to
offline
Unimportant/Frequent
4.209
3.181
.001
1.028
Polarizing
3.066
2.383
.001
0.683
Inappropriate
2.823
2.193
.001
0.630
Everyday Life
2.408
2.009
.002
0.399
Behavior
3.950
5.240
.001
-1.290
Change
2.774
3.371
.001
-0.597
Construct
Covariates p < .05
Everyday Life
None
Inappropriate
None
Polarizing
FS
p=.002 B=-.122
Unimportant/Frequent
FL
G
Y
p=.002 B=-.136
p=.022 B=-.412
p=.014 B=.181
Behavior
FS
G
p=.001 B=.183
p=.001 B=.497
Change
Y
A
p=.008 B=.158
p=.049 B=-.081
Covariates A: Age, G: Gender, FL: friend length, FS: Number of times friend seen in last year, Y:
Years in social networks.
7 Results
7.7
30
Final Model
A final model was generated based on the factor analysis of the questions about online and offline
behaviors (Figure 2 on page 31). The model is used to generate the hypotheses for in the following
section and to guide future research in the area of unfriending. The final model differs in several
ways from the initial model. Frequent posting is joined with posts about unimportant topics to
generate the construct frequent/unimportant posting. Politics and religion are separated from the
other inflammatory topics to generate the construct polarizing topics. Inflammatory topics, with
the removal of politics, religion and money is renamed inappropriate topics. Money joins the other
banal topics and is renamed everyday life topics. The new categories reflect the exploratory factor
analysis. The category names are based on an interpretation of the questions that remain in the
constructs. The offline behavior categories remain the same as in the initial model and are disliked
behavior and changes in the relationship.
7 Results
31
Fig. 2: Final Model
7 Results
7.8
32
Generated Hypotheses
The exploratory methods in this study have generated the following hypotheses that can be tested
in subsequent confirmatory research. These hypotheses are:
Friend Request Hypotheses
• H1a: Facebook users who initiate the friend request will be unfriended more often than
expected.
• H1b: Facebook users who receive the friend request will make more unfriending decisions
than expected.
• H1c: Facebook users will be more certain about who initiated the friend request when they
make the decision to unfriend compared to those who are unfriended.
Online/Offline Behavior Hypotheses
• H2a: Facebook users who unfriend for online reasons will agree at statistically significantly
higher levels that the person they unfriended posted about too often about unimportant topics.
• H2b: Facebook users who unfriend for online reasons will agree at statistically significantly
higher levels that the person they unfriended posted about polarizing topics.
• H2c: Facebook users who unfriend for online reasons will agree at statistically significantly
higher levels that the person they unfriended posted about inappropriate topics.
• H2d: Facebook users who unfriend for online reasons will agree at statistically significantly
higher levels that the person they unfriended posted about everyday life topics.
• H2e: Facebook users who unfriend for offline reasons will agree at statistically significantly
higher levels that they disliked the person’s behavior.
• H2f: Facebook users who unfriend for offline reasons will agree at statistically significantly
higher levels that they experienced a change in the relationship.
8 Limitations
8
33
Limitations
Sampling of this study is not random and therefore it would be difficult to say how the general
population perceives unfriending on Facebook. The survey respondents were recruited from Twitter and are likely to be different in many ways than the general Facebook population. Twitter
users were screened based on criteria that included the term “unfriend,” “defriend,” or “unfriending.” These users may be more likely to unfriend than others because they were posting about the
topic. Twitter users may be more tolerant of frequent/uninteresting posts compared to the general
population. This recruitment method was used because it showed that the person had an interest
in the unfriending phenomenon and may be more willing to take the survey as a result. The surveys may be biased toward technologically savvy users because of the large recruitment through
Twitter. Although the sample had a similar percentage of international responses (30.6%) to the
population of international users of Facebook (30%) the sample was from those who spoke English
and non-English speaking survey respondents are likely to be under-represented in this research.
The MANCOVA analysis used age, gender, education, etc. as covariates to eliminate intervallevel differences between these groups if differences are found. Most of the analysis did not find
statistically insignificant differences for the covariates when the results were adjusted.
The survey did not assess the role of privacy in unfriending behaviors. Personal privacy may be
a factor in unfriending decisions and is not used to evaluate unfriending behavior in this research.
Survey respondents were asked to choose whether they unfriended the person for that person’s
online or offline behaviors or to choose don’t know. Some survey respondents said that they did
not unfriend the person for their online or offline behavior but for a personal sense of privacy.
Some survey respondents abandoned the survey when they did not see options that fit their reason
for unfriending.
The survey did not assess the role of identity in unfriending behaviors. Several respondents
indicated that they had trouble maintaining a work identity and a personal identity in one user
account. Some choose to have two accounts that they manage with different types of friends and
9 Discussion and Conclusion
34
different status updates for the different identities. For example, some universities ask instructors
to have two accounts, one for students and one for their personal friends.
The survey relied on the memories of the survey respondents and did not measure who actually
initiated the friend request and who made the unfriending decision. The respondents memory
could be inaccurate and simple user data could be captured for this analysis that does not rely
on the survey respondents memory. Unfriending can occur for other reasons than the conscious
decision to unfriend a particular person; a person could have decided to deactivate their account.
9
Discussion and Conclusion
There are two major research questions that this study addresses.
1. What is the role of the friend request in unfriending decisions.
2. Can factors in unfriending decisions be found and do differences in the perception of online
and offline behaviors vary depending on the unfriending decision.
The research shows that survey respondents who initiate the friend request are unfriended more often than expected. Survey respondents who receive the friend request make more than the expected
number of unfriend decisions. Survey respondents were more certain of who initiated the request
when they did the unfriending compared to survey respondents who were being unfriended.
The requests can be summarized as follows:
• Survey Respondents who did the unfriending initiated the request 15% of the time. That is,
the survey respondent initiated the friend request and later dissolved the friendship.
• Survey Respondents who did the unfriending received the request 56% of the time. That is,
the survey respondent received the friend request and later dissolved the friendship.
Unfriending decisions are made for both offline and online reasons; more than half (57.0%) of
the survey respondents said they unfriended someone for online reasons vs. 26.9% for offline
10 Implications and Future Research
35
reasons with the remaining not sure. Those who unfriended for online reasons agreed more often
that the person they unfriended posted too frequently about unimportant topics. This construct
had the highest level of agreement for online post constructs and the largest difference of the
online constructs when compared to those who unfriended for offline reasons. The next highest
levels of agreement for unfriending for online reasons is posting about polarizing topics followed
by inappropriate topics. Everyday life posts had the lowest level of agreement that the person
unfriended posted too often about those topics.
Those who unfriended for offline reasons disliked the person’s behavior at a high level of agreement compared to those who unfriended for online reasons. This construct had the highest level of
agreement of all six constructs and had the largest difference when comparing against the unfriend
decision (online vs. offline). That is, many of the survey respondents agreed that they disliked
the person’s behavior that they unfriended; but those who unfriended for offline reasons disliked
the person at a statistically significantly higher level compared to those who unfriended for online
reasons. Survey respondents agreed that changes in the offline relationship (romantic relationship
dissolution, geographic change, etc.) occurred more often when they said they unfriended for
offline reasons compared to online reasons.
10
10.1
Implications and Future Research
Implications for Individual Users and Business Users
The unfriending behavior analyzed in this research has several implications for practice. Facebook
users who are negatively affected when unfriended may wish to change their posting behavior to
reduce the risks of being unfriended. Those who unfriend others for online reasons did so for the
following reasons in order: (1) the person posted too often about unimportant topics, (2) posted too
often about polarizing topics, (3) posted too often about inappropriate topics, and (4) posted too
often about everyday life. The implications to the individual user are discussed first and followed
by the implications to the business user.
10 Implications and Future Research
36
Users can limit the frequency of posts they make to the level of interest of their audience.
Facebook users may also try to limit their posts to interesting topics and eliminate those which
may be considered unimportant to their friends. Facebook allows users to make lists of people
to help narrowcast posts to interested users. Facebook users can use these lists to post to certain
groups of people who may be interested in a topic instead of all of their contacts. For example,
Facebook posts about local events (such as art openings, band concerts, Realtors who sell houses,
etc.) may be of little interest to those who live far away from the poster. Facebook users can make
a list of those who may be interested in local art galleries, for example, and send the post to only
those users. This may effectively reduce the noise in the news feed of Facebook users.
More specific topic areas for online unfriending are polarizing posts, inappropriate posts and
every day life posts. Polarizing topics (politics and religion) were the topic area that was most
strongly associated with those who unfriend for online reasons and these topics are likely to be a
problem for those who hold a different view of the poster’s political or religious view (although
this was not measured in this study). Users may wish to limit posting about these topics if the are
concerned about being unfriended or, again, may narrowcast their posts to those who may share
similar views or are more tolerant of these topics. Inappropriate topics (those which people find
inappropriate in some manner, sex, swearing, sexist, racist, and unflattering statements about Facebook users) are topics that Facebook users may already believe they are avoiding. Unlike political
and religious discussions where posters may knowingly and intentionally post about these polarizing topics those who post about inappropriate topics may not realize that the message received
by the recipient is interpreted as being “inappropriate.” For example, during the data collection
Arizona’s immigration law (SB 1070) was passed and in the news. Some Facebook users posted
about their support for the bill and some receivers of the posts interpreted the support as a racist
posting. Facebook users may want carefully consider whether their post may be interpreted by
others as inappropriate, whether they feel it is inappropriate or not, if they want to avoid being
unfriended. Every day life topics had the lowest levels of unfriending of the online reasons but
was still statistically significantly higher for those who said they unfriended for online reasons vs.
10 Implications and Future Research
37
offline reasons. The research indicates that all the posting constructs had statistically significantly
higher levels of agreement when the survey respondent said they unfriended someone for an online
reason compared to those who said they unfriended for an offline reason.
People who are not concerned about unfriending will not need to self-edit their posts any differently than they do today. Facebook users post about a variety of topics and do not need to limit their
speech in any way. The research indicates that some speech is more strongly related to unfriending
for online reasons than other speech. If people are negatively affected by unfriending they may
wish to limit their content to less polarizing and inappropriate speech and reduce the frequency of
their potentially unimportant posts.
Social media is used or planned to be used by 83% of recruiters, according to Jobvite’s 2010
report.15 The vast majority of the recruiters used LinkedIn (78.3%) which is designed as a social
network site for professionals; however, 54.6% also used Facebook to recruit. 58.1% of recruiters
said that they have successfully hired an employee through an online social network site. Recruiters
use social network sites to find information about the candidates. 70.3% said they occasionally or
always searched for a candidate’s profile with an additional 13.5% saying they searched when the
profile was provided by the candidate. Only 16.2% of the recruiters in the survey said they did
not review profiles at all. The report does not indicate what types of information recruiters seek
when they examine Facebook profiles but they may be looking for the same types of posts that
are associated with unfriending. Facebook users who are unfriended often may be limiting their
exposure to potential job opportunities because their network size is smaller. There is a balance
in the friends you might want to keep in a given social network; a Facebook user may want to
unfriend someone who posts inappropriately on the user’s wall so they are not associated with that
behavior if a recruiter does look at the profile. Facebook users should be aware that their posts and
the posts made by their friends may impact their future employment opportunities.
Facebook can be used by individuals and companies to promote their business. Some people
15
http://recruiting.jobvite.com/news/press-releases/pr/
jobvite-social-recruiting-survey-2010.php
10 Implications and Future Research
38
promote their business through their personal pages and others use business accounts (Facebook
prefers that companies use business accounts). Business users can promote their business on Facebook by having people like their business which in turn creates a link between the business and
the Facebook account. Facebook users who like a business are called fans of that business. Unlike
personal accounts where agreement is required between the parties to become friends, Facebook
users can become fans of a business unilaterally; that is, the business does not need to accept the
like request. Business users of Facebook may follow this same advice to avoid unfriending or
dissolution of the like link that connects an individual to a company. Businesses would like these
links to be persistent so they can send their marketing messages to its customers through this social
network channel. If we extend the research from individuals to businesses the following claims can
be made:
• Business users should avoid frequent and unimportant posts to their customers.
• Business that are not in the political in nature or religious in nature should probably refrain
from posting about these polarizing topics.
• Business users should refrain from posting inappropriate speech (sexist/racist/etc.).
10.2
Implications for Facebook
Facebook, the company, can provide mechanisms that create a better user experience to reduce
the amount of unfriending on their social network. Facebook’s mission is to, “give people the
power to share and make the world more open and connected.”16 Facebook’s goal is to allow its
users to make more connections not reduce the number of connections so they may want to limit
the amount of unfriending on their network. Facebook actively encourages its users to find new
friends by placing a “find friends” button on the user’s news feed which will launch an application
to find friends from existing contacts. Facebook also suggests that people reconnect with others
in their network by making suggestions to its users to send messages to those who have not been
16
http://www.facebook.com/facebook?v=info
10 Implications and Future Research
39
contacted for a period of time. Facebook allows its users to determine which Facebook users
have joined the network because the user sent a friend request to a person who was not yet a
Facebook member. Facebook also makes it difficult to unfriend people from the network. There is
no way to mass unfriend; a user must go to each individual’s page, scroll down and click the link
“Remove from Friends” to unfriend. This link is placed in a relatively obscure location compared
to the prominently placed “Add as Friend” button that is displayed right next to a person’s name.
Facebook’s mission and features seem to indicate that Facebook wants a more connected user
community and does not desire connections to be dissolved.
Facebook could make a variety of changes to potentially reduce unfriending. Facebook could
bundle posts that frequent posters make so that the receiver’s news feed is not dominated by the
most frequent posters. That is, if person x makes ten posts a day then Facebook could put all
of person x’s posts in a single group instead placing the posts in chronological order in the news
feed. This option should be configurable by the end user to allow the option to bundle or not
bundle based on preferences. Facebook could make generating and maintaining user lists easier
to use. Facebook users appear to be relatively unaware of these features or are not using them
to a great extent. To manage different social identities some Facebook users resort to generating
multiple user accounts to manage their posts instead of using the list feature. Some universities ask
instructors at their institution to maintain two different Facebook accounts one for their social life
and one that their students can friend (Young, 2010). Facebook could determine whether a post is
geographically local in nature through semantic measures (local events as real estate information,
posts about musical acts who are playing at a particular venue, etc.) and suggest to the user that
they limit the recipients list to those who are in close proximity.
Facebook users have the option to hide a person y if the content of person y’s posts does not
appeal to the end user instead of directly unfriending the person. Facebook also provides the option
to hide certain application updates, such as game updates, without hiding the person completely.
These mechanism are a bit heavy-handed and do not allow for more nuance that allows users to
see posts that they want to see and limit posts that they do not want to see. Facebook could provide
10 Implications and Future Research
40
machine learning mechanisms to allow end users to train the news feed to show posts that the user
finds interesting and hide posts that are of little interest. Users could rate their level of interest of a
post in a 1 through 5 rating system or a simple rating system where they indicate they would like
to see more or fewer posts that are similar to the post shown. As the algorithm is trained over time
Facebook users may limit the amount of uninteresting posts in their news feed and create a better
user experience.
10.3
Future Research
There are many areas in which this research on unfriending can be extended and address some of
the limitations of this research. This research looks at unfriendings that have already occurred and
asks whether they occurred for an online/offline reason. A longitudinal study could be developed
with participants to examine the social network as it exists and see how it changes over time with
attention to unfriending decisions. In particular, a probability function (logistic regression) could
be developed to determine what kinds of posts and behaviors will likely lead to unfriending. The
research can extend beyond unfriending and examine the role that hiding and blocking a person
has on a social network. This research looked at online and offline reasons for unfriending; future research can capture the role of identity and privacy in unfriending decisions. This study was
exploratory in nature but future confirmatory work can be completed that use the hypotheses generated in this research. Based on the constructs found in this research, more specific questions in the
construct areas (frequent/unimportant posts, polarizing topics, inappropriate topics and everyday
life topics) could be generated that would yield more reliable constructs. Future research could
have a random sample of Facebook users instead of relying on a convenience sample of Twitter
users. Some survey respondents stated that they mainly used Facebook for business purposes;
these users were interested in the business implications of unfriending on Facebook which future
work could address.
Finally, stronger theory can be applied to explain why people unfriend. Gift-giving is a theory
10 Implications and Future Research
41
that may help explain social behavior in online networks (Skageby, 2010; Lampel and Bhalla,
2007; Burnett, 2011). Unfriending behaviors may be examined using a similar theoretical frame.
Gift-giving has been used to explain why members of an online community contribute posts (such
as product reviews) to a forum (Lampel and Bhalla, 2007). Gift-giving is a combination of a
rational calculation of costs and benefits with potential reciprocity in the near future and driven by
a desire to contribute to the general welfare (altruism) or repay past generosity. Neither altruism or
reciprocity (for a like gift) independently provides enough motivation to explain the contributions
users make to a network. Lampel and Bhalla (2007) argue that these gifts are not simply given
for altruistic reasons or the expectation of reciprocity but the desire to increase the contributor’s
status in the network. In this case, the reciprocity may be more immediately received; that is, the
person making the contribution may not be seeking a reciprocal response for information that they
seek but an increase in their personal status. In the case of a social network the gift may be the
initiation of the friend request. Given that, in the real world, the initiator of the request tends to
have lower status than the receiver (Hallinan, 1979) the initiator may be seeking an increase of
their status in the network. It is possible that the initiator gives the gift of the friendship and the
receives an increase of status. This research shows that the recipient of the friend request makes
more than the expected number of unfriending decisions; that is, they accept the friend request
and later unfriend more than expected number of times. Therefore, in the case where the recipient
unfriends the initiator, the recipient may be rejecting the gift of the person’s friendship (which they
may not value) and, in turn, lower the status of the initiator. It may also be that the recipient of the
request did not want to be associated with a lower status person and therefore rejects the friendship
in order to increase their own status.
Social network participants can not simply be trying to gather as many friends as possible to
increase their social capital - if that were the case then no one would unfriend. Just as friendship
dissolution is not simply formation in reverse, care must be taken when applying a theory for online
friendship formation to ensure it also explains the dissolution of the relationship. Ideally, theories
about connections on social networks need to account for both friending and unfriending behavior.
10 Implications and Future Research
42
At present, most research on social networks examines friendship formation on social networks
and does not examine the entire life-cycle which may include friendship dissolution. Generating
a theory for a full life cycle of online social network friendship that includes both formation and
dissolution needs further development.
This research shows how the friend request and online and offline behaviors contribute to unfriending decisions. There are many areas in which this research can be extended and improved to
generate new knowledge about how users navigate the online world.
10 Implications and Future Research
43
References
Baxter, L. A. (1979). Self-disclosure as a relationship disengagement strategy: An exploratory investigation. Human Communication Research, 5:215–222.
Baym, N. K. (2010). Personal Connections in the Digital Age. Digital Media and Society. Polity.
Baym, N. K., Zhang, Y. B., and Lin, M.-C. (2004). Social interactions across media : Interpersonal communication on teh internet,
telephone and face-to-face. New Media and Society, 6(3):299–318.
Bernstein, E. (2009). How facebook ruins friendships. The Wall Street Journal, page Life & Style.
Blanchard, A. (2004). Virtual behavior settings: An application of behavior setting theories to virtual communities. Journal of
Computer-Mediated Communication, 9(2).
boyd, D. M. and Ellison, N. B. (2007). Social network sites: Definition, history and scholarship. Journal of Computer-Mediated
Communication, 13(1):1.
Burnett, R. (2011). The Handbook of Internet Studies. Wiley-Blackwell.
Bushman, B. B. and Holt-Lundstad, J. (2009). Understanding social relationship maintenance among friends: Why we don’t end
those furstrating friendships. Journal of Social and Clinical Psychology, 28:749–778.
Chua, C. E. H. (2009). Why do virtual communities regulate speech? Communication Monographs, 76(2):234–261.
Dillman, D. A. (1978). Mail and Telephone Surveys: The Total Design Method. Wiley.
Duck, S. W. (1982). Personal Relationships and Personal Constructs: A Study of Friendship Formation. John Wiley.
Dwyer, C., Hiltz, S. R., and Passerini, K. (2007). Trust and privacy concern within social networking sites: A comparison of
facebook and myspace. In Proceedings of the Thirteenth Americas Conference on Information Systems, Proceedings of the
Thirteenth Americas Conference on Information Systems.
Ellison, N. B., Steinfield, C., and Lampe, C. (2007). The benefits of facebook "friends:" social capital and college students’ use of
online social network sites. Journal of Computer-Mediated Communication, 12:1143–1168.
George, D. and Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference. 11.0 update. Allyn & Bacon,
Boston, 4th edition.
Gilbert, E. and Karahalios, K. (2009). Predicting tie strength with social media. Boston, MA. CHI.
Goldsmith, B. (2009). "unfriend" named word of 2009. Reuters, 1:1.
Grimes, W. (1992). Computer networks foster cultural chatting for modern times. NY Times, pages B1, B4.
Gross, R. and Acquisti, A. (2005). Information revelation and privacy in online social networks. In Proceedings of the 2005 ACM
workshop on Privacy in the electronic society, volume 1, pages 71–80.
Hallinan, M. T. (1979). The process of friendship formation. Social Networks, 1(2):192–210.
Hallinan, M. T. and Williams, R. A. (1987). The stability of students’ interracial friendships. American Sociological Review,
52(5):652–664.
Hartup, W. W. and Stevens, N. (1999). Friendships and adaptation across the life span. Current Directions in Psychological Science,
8(3):76–79.
Haythornwaite, C. (2002). Strong, weak, and latent ties and the impact of new media. The Information Society, 18(5):385–401.
Hewitt, A. and Forte, A. (2006). Crossing boundaries: Identity management and student/faculty relationships on the facebook.
Poster presented at CSCW, Banff, Alberta, Canada.
Jaeger, R. G. and Halliday, T. R. (1998). On confirmatory versus exploratory research. Herpetologica, 54:S64–S66.
Joreskog, K. (1969). A general approach to confirmatory maximum likelihood factor analysis. Psychometrika, 34(2):183–202.
Joseph F. Hair, J., Black, W. C., Babin, B. J., Anderson, R. E., and Tatham, R. L. (2006). Multivariate Data Analysis. Pearson
Education, Inc., 6th edition.
Lampel, J. and Bhalla, A. (2007). The role of status seeking in online communities: Giving the gift of experience. Journal of
Computer-Mediated Communication, 12(2):434–455.
Lea, M. and Duck, S. (1982). A model for the role of similarity of values in frienship development. British Journal of Social
Psyschology, 21:301–310.
Leech, N., Barrett, K., and Morgan, G. (2008). SPSS for intermediate statistics: Use and interpretation. Erlbaum/Taylor & Francis
Group., 3rd edition.
Levinger, G. (195). Marital cohesiveness and dissolution: An integrative review. Journal of Marriage and the Family, 27(2):19–28.
McPherson, M., Smith-Lovin, L., and Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annaul Review of
Sociology, 27:415–444.
Metts, S. (1994). The Dark side of interpersonal communication. Routledge.
10 Implications and Future Research
44
Nielsen, L. B. (2002). Subtle, pervasive, harmful: Racist and sexist remarks in public as hate speech. Journal of Social Issues,
58(2):265–280.
Nuttal, C. and Gelles, D. (2010). Facebook becomes bigger hit than google. Financial Times.
Oliver, R. T. (1961). Conversational rules - their use and abuse. Communication Quarterly, 9(2):19–22.
Owens, R. A. (2003). Friendship Features Associated wtih College Students’ Friendship Maintenance and Dissolution Following
Problems. PhD thesis, West Virginia University.
Parks, M. R. and Floyd, K. (1996). Making friends in cyberspace. Journal of Communication, 46(1):80–96.
Randaza, J. (2009). Go Tweet Yourself. Adams Media.
Raphael, J. (2009). Facebook overtakes myspace in u.s. PC World.
Riva, G. (2001). Say not to Say: New perspectives on miscommunication. IOS Press.
Skageby, J. (2010). Gift-giving as a conceptual framework: framing social behavior in online networks. Journal of Information
Technology, 25(2):170–177.
Smith, C. B., McLaughlin, M. L., and Osborne, K. K. (1997). Conduct control on usenet. Journal of Computer-Mediated Communication, 2(4).
Sprecher, S. and Fehr, B. (1998). The dissolution of close relationships. Edwards Brother.
Sternberg, J. (2000). Virtual misbehavaior: Breaking rules of conduct in online environements. In Proceedings of the Media
Ecology Association, volume 1, pages 53–60.
Stutzman, F. (2006). An evaluation of identity-sharing behavior in social network communities. Journal of the International Digital
Media and Arts Association, 3(1):10–18.
Walther, J. B. (1992). A longitudinal experiment on relationahl tone in computer-mediated and face to face interaction. Proceedings
of the Hawaii International Conference on System Sciences, 4:220–231.
Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23(1):3–43.
Wang, S. S., Moon, S.-I., Kwon, K. H., Evans, C. A., and Stefanone, M. A. (2010). Face off: Implications of visual cues on
initiating friendship on facebook. Computers in Human Behavior, 26:226–234.
Wright, P. (1984). Self-referent motiviation and the intrisic quality of friendship. Journal of Social and Personal Relationships,
1:115–30.
Yates, J. and Orlikowski, W. J. (1992). Genres of organizational communication: A structurational approach to studying communication and media. The Academy of Management Review, 17(2):299–326.
Young, J. R. (2010). How social networking helps teaching (and worries some professors). The Chronicle of Higher Education,
page online.