University of Southern California Dissertations and Theses

Why our audience share? Improving social media effectiveness using experiments

(USC Thesis Other)

Why our audience share? Improving social media effectiveness using experiments

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

WHY OUR AUDIENCE SHARE? IMPROVING
SOCIAL MEDIA EFFECTIVENESS USING
EXPERIMENTS
by
Jing Xu
A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF ARTS
(STRATEGIC PUBLIC RELATIONS)
May 2015
Copyright 2015
Jing Xu
1
Acknowledgements
I truly appreciate the time professors, experiment participants spent to help me conduct the
experiments and accomplish this thesis. This research is a trail to translate academic theories
to practitioners, to apply academic research method into social media practice. During this
process, my thesis committee gave me tremendous help. I would like to thank my thesis
committee chair, Kjerstin Thorson. Her enlightening directions and supportive words have
always encouraged me to overcome the challenges in developing this paper. I would also
thank my committee members Brenda Lynch and Matthew Le Veque for their insightful
suggestions and guidance on my thesis.
2
Abstract
Most practitioners have realized the essential role social media could play in
communication, marketing and issue advocacy. But few of them are confident about how to
use social media effectively. This paper tries to address this issue in two ways: theoretically
and methodologically. First, this paper examines previous research, to reveal the mechanism
about why people share and what people like to share. In general, self-serving factors act as
internal motivations; content factors may further trigger people’s interests; context factors
could moderate people’s willingness and preference of sharing. This theory cluster provides a
pool of potential factors that may increase social media effectiveness. Second, this paper
argues that controlled experiments could serve as an effective, low-cost way for practitioners
to learn their audience and test which potential factors would work on their social media
accounts. Compared to other existing methods, controlled experiments help to reveal the
underlying factors of social transmission in a reliable way. The researcher also conducts three
sample experiments and creates a workflow as an actionable guidance on how to conduct
controlled experiments in practice. This paper aims to help practitioners conceptually
understand the mechanism about how their audience share social media message. Only by
deeply learning their audience, can practitioners be proactive and design customized, sharable,
effective social media messages.
3
Table of Contents
Acknowledgements .......................................................................................................................... 1
Abstract ................................................................................................................................................... 2
List of Tables .......................................................................................................................................... 4
I. Opportunity and Challenge: How to Design a Successful Social Media Campaign on Our Own
Channel? ................................................................................................................................................. 5
II. Theoretical Review: How Do People Make Decisions on Which Social Message to Share?......... 11
2.1 Interpersonal Communication Perspective: Self-serving Factors ............................................ 13
2.2 Content Perspective: Inherent factors of messages ................................................................ 17
2.3 Network Perspective: Connection with Others .................................................................... 25
III. Experiment: Test Message before the Campaign ..................................................................... 28
3.1 The Adoption of Experiments: Seldom Used in Communication Practice............................... 28
3.2 Message Testing....................................................................................................................... 29
3.3 Social Media Message Testing: Begin to Apply Experimental Method in Social Media .......... 32
IV. Sample Experiments ................................................................................................................. 35
4.1 Study One: Practical Utility and Social Media Message Effectiveness .................................... 36
4.2 Study Two: Emotion Arousal and Social Media Message Effectiveness .................................. 41
4.3 Study Three: Intersection Effect of Practical Utility and Emotion Arousal on Social Media
Message Effectiveness ...................................................................................................................... 47
V. Practical Implications .................................................................................................................... 51
5.1 When Can Experiments Be Used? ............................................................................................ 51
5.2 Why Are Experiments Helpful? ................................................................................................ 52
5.3 How to Conduct Experiments .................................................................................................. 55
VI. Conclusion ................................................................................................................................. 68
References ............................................................................................................................................ 69
Appendix: Consent Form Used in the Sample Experiments ........................................................... 74

4
List of Figures
Figure 1: How Do people Make Decisions on Sharing ...........................................................13
Figure 2: Relationship between Controversy and Conversation in a Field Study ...................23
Figure 3: A/B Test of comScore...............................................................................................31
Figure 4: Brand Profiles created for Green Market and Pacific Airline ..................................36
Figure 5: Tweets Used in Study One .......................................................................................39
Figure 6: Tweets Used in Study Two .......................................................................................44
Figure 7: Workflow of Conducting Experiments in Practical Context....................................55
Figure 8: Operationalization of People’s Willingness to Share ...............................................59
Figure 9: Basic Experiment Design: One Independent Variable .............................................61
Figure 10: Basic Experiment Design: One Independent Variable without Pretest ..................61
Figure 11: Basic Experiment Design: One Independent Variable—Example .........................62
Figure 12: Basic Experiment Design: Two Independent Variables .........................................63
Figure 13: Basic Experiment Design: Two Independent Variables—Example .......................63
List of Tables
Table 1: Examples of High-arousal Emotions and Low-arousal Emotions.............................20
Table 2: List of Informational Value Factors ...........................................................................22
Table 3: Means of High- and Low- practical Utility Group: Green Market............................40
Table 4: Means of High- and Low- practical Utility Group: Pacific Airline ...........................41
Table 5: Means of High- and Low- emotion Arousal Group: Green Market...........................46
Table 6: Means of High- and Low- emotion Arousal Group: Pacific Airline..........................46
5
I. Opportunity and Challenge: How to Design a Successful Social Media Campaign
on Our Own Channel?
Social media has deeply transformed the way organizations communicate with their
audience. It allows organizations to publish their information in real-time, to listen what
people are concerned about, to jump into a popular conversation, and to engage the public in
their own storytelling. With social media, organizations are able to directly interact with
people who used to be hard to reach.
Most practitioners have realized the power of social media and adopted it into practice.
A survey from Social Media Examiner (Stein 2014) suggested that in 2014, the majority
(97%) of marketers stated that they were participating in social media and “a significant 92%
of marketers believed that social media is important to their business.”
With the rise of social media, some campaigns have stood out from others and achieve
great success. On social media, messages of these campaigns spread widely within a short
time, attracting huge amount of attention and gaining support nationally and globally. A look
at some of the most viral campaigns in the history shows the impact social media could have
on communicating, marketing and issue advocacy.
Kony 2012
One successful example is the viral video called “Kony 2012.” In 2012, this 30-minute
documentary, which was shot by a nonprofit organization called Invisible Children, was
6
released on social media, aiming to call for help to stop the crimes of Joseph Kony, the head
of the Lord’s Resistance Army, and to raise people’s awareness about this issue. The video
quickly went viral and made Joseph Kony “the web’s most hated man” (Orden and Bariyo
2014). Within eight days of its releasing, the video had been viewed more than 76 million
times on YouTube and 16 million times on Vimeo, and there were about 5 million related
tweets posted (Rainie et al. 2012). Mainstream media, including all three evening network
newscasts, the New York Times, the Wall Street Journal, etc., also covered this video and
expressed their opinions on this campaign.
Social media helped to turn this cause into a widely discussed topic within a short time
period. According to a Pew Research Center report (Rainie et al. 2012), 58% of young adults
(aged 18 – 29) heard of this video, among which, 62% learned about it either from social
media or other Internet sources. Fifty percent of those aged 30 – 49 heard about the video,
among which, 44% learned about it either from social media or other Internet sources.
The success of this campaign is not an accident. It understood how to catch the public’s
attention and motivate them to take action. Instead of using cold facts or numbers, the
campaign illustrates Kony’s crimes by telling an emotional story. In the video, Jacob, a young
Ugandan with tears and despair in his eyes, tells us how his brother was killed by the Lord’s
Resistance Army. The video also tried to include us in the story. An analogy was made
between the filmmaker’s son and Jacob. “Every single person in the world started in this way,”
says a voiceover accompanying footage showing his baby’s birth, “he didn’t choose where or
when he was born, but because he’s here, he matters.” By building this connection, we
7
became part of Jacob’s story. People, especially young people, were shocked and appalled
when they saw a person similar to them suffering from Kony’s crimes (Suddath 2012; Preston
2012).
The video then provided a way for viewers to join the action of catching Kony, by
sharing the video or donating money, to put Kony’s name everywhere. Moved by this story,
millions of people pressed the share button and donated money in response to the video’s call.
Ice Bucket Challenge
Another recent example is the Ice Bucket Challenge, in 2014, in support of research for
Amyotrophic Lateral Sclerosis, or ALS. The challenge requires people to either pour a bucket
of ice water on themselves or donate $100 to the ALS Association, while naming three of
their friends on Facebook to take this challenge next.
The social media network turned a vision of a group of people into a global action.
Hundreds of thousands of people joined this campaign, including politicians like George W.
Bush, tech titans like Mark Zuckerberg and Bill Gates, and other celebrities. The public
began to learn about ALS from this campaign. According to the Wikimedia Foundation, the
“ALS Wikipedia page in English was viewed 2.89 million times in August” after the
campaign, marking an 18-fold increase of the previous number of views (Petronzio 2014).
Within one month, over $100 million was raised for the ALS Association (ALS Association
2014).
As with Kony 2012, the Ice Bucket Challenge managed to harness what people like to
8
share on social media. The campaign combined a good cause with a fun challenge (Stenovec
2014; Boko 2014). Pouring a bucket of ice water over your head or watching others doing it
is somewhat comical, turning the idea of helping ALS patients into a tangible, easy-to-do
action. In addition, the challenge requires people to nominate three friends. This “pass it on”
feature is in accord with the nature of social media networks. As Stenovec (2014) pointed out,
the campaign itself is inherently spreadable.
In contrast, many social media campaigns failed to achieve expected goals: the posts
received few responses and were soon drowned out by other conversations. What makes a
social media success? Can we duplicate the success of Kony 2012 or the Ice Bucket
Challenge?
This becomes a challenge for most practitioners. While communication or marketing
professionals understand the necessity of social/digital media adoption and witness many
successful examples, few are confident about how to use their own social/digital media
effectively. A survey conducted by Adobe shows that marketers are unconfident in their
digital ability, and “only 48% of digital marketers feel highly proficient in digital marketing”
(Adobe 2013). According to the 2014 survey by Stein, the number one concern of marketers
(91%) is “which social tactics work best.”
Therefore, the main purpose of this paper is to help practitioners improve the social
media effectiveness of their organizations.
First, this paper discusses how people make decisions on sharing social messages,
9
aiming to help practitioners understand the factors which might trigger people’s interests in
social transmission. The cases introduced above provided clues about what could make a
successful social media campaign: emotional story, call to action, good cause, amusement
factor, etc. Are they the common rules behind success? How do those potential factors take
effect? What is the mechanism of content going viral? By examining previous research and
theories, this paper aims to provide a knowledge system about why people share and what
they like to share, to form a solid base for practitioners to design social media strategies.
Q1: How do people make decisions on which social messages to share?
After theoretically understanding the influential factors, this paper provides a method to
apply this knowledge to social media practice: controlled experiments. People always learn
from previous cases or others’ advice. This paper, however, argues that experimentation could
serve as an effective way for practitioners to learn about their audience.
This idea is enlightened by the testing method used by some communication campaigns.
For instance, during the presidential race of 2012, Obama’s team used A/B testing on email
message, to find out the most effective email title, subject and formatting for gaining the
public’s support. The team sent out numerous draft emails to small groups of supporters. The
variables tested included subject line, amount of money they would ask people for, wording,
email formatting, etc. The team would select the winning email after testing and send it to
millions of subscribers. According to their testing data, the most effective subject line in June
was “I will be outspent,” raising more than $2.5 million. The email-message testing was one
of the keys to the success of Obama’s campaign. As Bloomberg reported (Green 2012),
10
“Most of the $690 million Obama raised online came from fundraising e-mails.”
This paper explores how to adopt the controlled experiments, which are more rigorous
than testing conducted by Obama campaign, into social media practice. Three sample
experiments are conducted to provide practitioners with an example of how to use this
method in a practical context. The advantages and feasibility of practical social media
experimentation will then be discussed. Finally, this paper creates a workflow based on the
samples to provide detailed guidance on experiment design and operation
Q2: How can practitioners use experiments as a prelaunch research method to
develop customized, effective social media messages?
11
II. Theoretical Review: How Do People Make Decisions on Which Social Message to
Share?
To improve social media effectiveness, practitioners need to understand what social
media effectiveness entails.
According to the International Association for the Measurement and Evaluation of
Communication (AMEC 2014), social media effort can be evaluated through the phases
outlined below. This framework was developed from 15 models used by Forrester, McKinsey,
Ketchum and other companies.
Exposure: Content andmessage reach to potential audience
Engagement: Interaction that occurs in response to content, such as people engaging
with you through your account, or talking about you on their own channel
Influence: Ability to cause or contribute to a change in opinion or behavior
Impact: Effect on the target audience, including financial value
Advocacy: Act of pleading or making the case for something, such as a
recommendation or a call to action
This paper suggests that understanding how people make decisions about sharing, such
as why they share and what they like to share, is essential for improving social media
effectiveness.
First, understanding the mechanism of social transmission helps practitioners to learn
about their audience. While many brands have begun to listen to what people are talking
12
about and tried to follow the trends on social media, few of them understand why people are
willing to share specific topics. Only by learning about what motivates people to share,
however, can brands be proactive and create more targeted, sharable and effective social
media contents.
Second, sharing is the starting point for social media to take effect. Sharability is one of
the key features through which social media differ from pre-internet media. When a brand
posts a message, only if people share it with their friends can this message have the chance to
reach a large audience and engage with them. In AMEC’s framework, exposure is the
fundament for a large number of engagements, wide impact and the success of advocacy.
Therefore, this paper believes that understanding the psychological process of sharing and
increasing the likelihood for a message to be shared are essential for improving social media
effectiveness.
Third, sharing itself could serve as a social mobilization for people to take action (Bond
et al. 2012). People’s sharing may encourage their friends to share or take action the message
advocates.
This paper examined current research, trying to help practitioners understand how
people make decisions about sharing messages at a micro-level in a general communication
context.
13
Figure 1: How Do People Make Decisions on Sharing?
2.1 Interpersonal Communication Perspective: Self-serving Factors
1) Impression Management
One reason for people to share is to shape how others think of them. In general, people
tend to share messages which could make them look good, while avoiding messages which
may leave others with a negative impression (Berger 2014, 588-590; Hennig-Thurau et al.
2004. 49-50).
This might be one of the reasons for the Ice Bucket Challenge going viral. The campaign
has a good cause: helping ALS patients. By sharing the Ice Bucket Challenge or taking part in
the activities, people have the chance to show their kindness and sense of social responsibility
to others. In an email to the Huffington Post, Jonah Berger wrote that “ALS is a good cause,
so when someone asks you directly to do this, it’s hard to turn them down without seeming
like a bad person” (Stenovec, 2014).
People also care about what type of image they show to others. This may influence their
Why do People Share?
What do People Share?
What motivates
people internally?
What types of content
trigger people’s interest?
How do others influence
people’s decisions?
Self-serving Factors
Content Factors
Network Factors
14
selection about what to share. For example, if a person expects others to consider him/her as a
technology expert, he/she may post things about technology news, trends, etc. In other words,
people like to share particular topics to display their interests, expertise, or character (Berger
2014, 589-590; Chung and Darke 2006, 276-277).
2) Emotion Regulation
People may communicate with others for emotion regulation, which is defined by Gross
(1998, 275) as “the ways people manage which emotions they have, when they have them,
and how they experience and express them” (Cited by Berger 2014, 592).
First, people seek social support, such as comfort and consolation, through social
sharing, especially after some negative experience (Rimé 2007; Berger 2014, 592). Second,
people talk to others to vent their emotion (Hennig-Thurau et al. 2004, 47-48; Berger 2014,
592). When suffering some negative emotion, people are probably eager to have someone to
talk to, instead of keeping the emotion inside. Zech’s study (1999) showed that “90% of
people believe that sharing an emotional experience will be relieving” (Cited by Berger 2014,
592). On the other hand, people may share their positive experience because it helps to relive
the pleasant feeling (Hennig-Thurau et al. 2004, 47-48; Berger 2014, 593). Langston (1994)
even suggested that “communicating positive events to others enhanced positive affect, even
above and beyond the affect associated with the experiences itself.” (Cited by Berger 2014,
593) In addition, communicating with others helps people to understand what is happening,
why it is happening (Rimé 2009, 80-81; Berger 2014, 592) and to confirm their feeling and
15
judgment (Dichter 1966, Cited by Berger 2014, 592). Researchers found that consumers
talked to others to gain support for their decision even after they have made a decision
(Berger 2014, 592).
3) Information Acquisition
Interpersonal communication also serves as a way for people to obtain information.
People often ask others for advice or solutions, especially when the decision they have to
make is “risky, important, complex, or uncertainty-ridden” or there is little “trustworthy
information” (Berger 2014, 594). This is probably why we can always see posts like “how to
choose bewteen the Canon SX50 and Canon SX40” or “should I buy that car”, and why many
of us rely on Yelp to choose an eating place. By talking to others, people hope to gain more
trustworthy information and opinions to reduce the risk, increase the certainty and thus make
a wise decision, or confirm the decision they have already made (Hennig-Thurau et al. 2004,
47-48; Sundaram, Mitra, & Webster, 1998; Berger 2014, 594). An empirical study conducted
by Lovett, Peres, and Shachar (2013, 438) could support this point. According to their
analysis over more than 600 brands, the ones with higher perceived risk are more likely to be
talked about.
4) Social Bonding
People also talk to others to reinforce social bonding. Human beings have the desire to
connect with others (Baumeister & Leary 1995; Cited by Berger 2014, 595) and to be part of
16
the group. We talk to, text, or email others to show we care about them. We discuss a topic
with friends to display we have something in common. Ritson and Elliott’s study (1999, Cited
by Berger 2014, 595) showed that talking about popular advertisements gives teenagers “a
type of social currency that allows them to fit in with their peers.” Similarly, people talk
about a brand or product, such as a luxury good, to show they are in the community with
“like-minded others” (Muniz and O'Guinn 2001; Cited by Berger 2014, 595).
Communicating with others not only helps to reinforce shared views and group
membership, but also reduces loneliness and the feeling of social exclusion (Berger 2014,
595). This may lead people to share topics related to common concerns, such as the weather,
traffic, famous stars, or topics that easily arouse others’ emotions, to resonate with their peers
and strengthen the connection.
5) Persuading Others
Finally, people may share words to persuade others. For example, we may describe how
good the food is, in the hope our friends agree with us or we can have dinner together there.
Kirchler’s research (1993, 417-419) showed that spouses use different tactics, such as
reasoning, emotion, bargaining, etc., to persuade or convince their partners when making
purchasing decisions. To achieve the expected persuasive effect, people may tend to share
information with polarized emotion (extremely bad or good) or emotion that can stimulate
others to take action (Berger 2014, 596).
17
2.2 Content Perspective: Inherent factors of messages
The factors above not only influence whether people are willing to share, but also affect
what content people tend to share (Berger, 2014). For example, to leave a knowledgeable
impression on others, people may tend to share a message with useful information. Several
studies have explored what specific content factors are more likely to trigger people’s
willingness to share.
1) Practical Utility
People like to share messages that contain useful information, such as discounts, advice,
and reviews (Berger and Milkman 2012, 193). In Berger and Milkman’s study (2012, 197),
for example, informative stories like restaurant reviews are more likely to make the New York
Times’ most e-mailed list. This is probably because people hope to leave others with the
impression that they are smart and helpful (Berger 2014, 590), or because useful information
has “social exchange value” (Homans 1958; cited by Berger and Milkman 2012, 193).
Many social media campaigns featuring practical utility have gained success. For
instance, Chipotle conducted its Halloween campaign called “Boorito” in 2012. It tweeted
that customers could buy a burrito, bowl, salad or tacos for just $2 if they dressed in costume
and visited a Chipotle restaurant on Halloween. The money would be used for charity. This
post soon spread across Twitter. Within one week, Chipotle reached a total volume of about
189,000 on Twitter and 348,000 comments on this campaign (Ives, 2012).
18
2) Emotion
Scholars discuss emotional factors in two ways: emotion valence and emotion arousal.
a. Emotion valence: positive vs. negative
In psychology, valence refers to the intrinsic attractiveness (positive valence) or
averseness (negative valence) of an object or situation. People’s preference for emotion
valence, positive or negative, varies in different situations.
People tend to share positive content when they want to show expertise (Wojnicki and
Godes 2011, cited by Berger 2014, 591), associate themselves with positive things and leave
others with a good impression. In addition, as discussed in the emotion regulation part above,
people may share a positive experience to relive pleasant feelings. An empirical example for
this notion is the field experiment conducted by Berger and Milkman (2012). After examining
7,000 New York Times’ articles, they found that while positive or negative content “is more
viral than content that does not evoke emotion, positive content is more viral than negative
content”.
Negative content, on the other hand, is more likely to be shared when people want to
manage their negative emotions (Berger 2014, 593), such as venting their anxiety to seek
comfort and consolation. This is consistent with people’s need for emotion regulation, as
mentioned above. When suffering from some negative experience or emotion, talking to
others may make them feel better and find consolation.
19
b. Emotion arousal: high arousal vs. low arousal
“Arousal is a state of mobilization” (Berger and Milkman 2012, 193). Emotion arousal
reflects the extent to which the emotion can stimulate people to take action. Physiologically,
high arousal encourages people to take action, while low arousal features relaxation and
deactivation. Smith and Ellsworth (1985) suggested that emotion can be differentiated based
on the arousal level they evoke. For instance, anger is a high-arousal emotion and sadness can
be regarded as a low-arousal emotion, though both of them belong to negative emotion.
Berger and Milkman (2012) pointed out that there is a causal impact between emotional
arousal and virality. The way emotion affects transmission is more complex than valence
alone. According to their series of experiments, messages with a higher arousal level were
more likely to be shared. Participants presented with an amusing ad were more willing to
share the content than those presented with a less amusing ad (with other factors of the two
ads being the same). Participants were more likely to share a story that made them angry than
a story that induced less anger (with other factors of the two stories being the same). In
contrast, contents that evoke emotion with a lower arousal level are less likely to be shared.
Participants who read the high-sad version of the news are less likely to share than those who
read the low-sad version of the news (with other factors of the two news being the same)
(Berger and Milkman, 2012).
While many practitioners adopt positive high-arousal emotion, such as amusement, in
their campaigns, some practitioners have tried to use negative high-arousal emotion to trigger
discussion. One recent example is Nationwide’s commercial aired during 2015 Super Bowl. A
20
young boy in the ad listed all the things he cannot do—such as traveling or having a
girlfriend—because he has already died in a preventable accident. “I cannot grow up, because
I died from an accident.” The depressing, strong negative emotion shocked the public. The
company was mentioned more than 234,810 times on social media during the game, making
itself the most mentioned Super Bowl advertiser (O’Reilly, 2015). When creating the ad,
Nationwide prepared two advertising narratives—the depressing died boy story and another
regular story—and conducted focus group meetings to compare people’s reaction towards the
two ads. According to its Chief Marketing Officer Matt Jauchius, the young boy’s story
received better results, and they anticipated this ad would cause a conversation. Though there
are lots of negative comments in the conversation and it might be risky for the company, this
advertisement shows an example of how an organization can use negative high-arousal
emotion and its potential viral effect.
Table 1: Examples of High-arousal Emotions and Low-arousal Emotions
High-arousal Emotion amusement, excitement, surprise, novel, anxiety, anger…
Low-arousal Emotion sadness…
One reason for arousal’s effect would be emotion regulation (Berger 2014, 593). When
people experienced high-arousal emotion, which is characterized by activity, they might have
a greater desire to vent or to rehash their emotions with others. Another reason is to persuade
others (Berger 2014, 596). When people plan to persuade others, they may convey
high-arousal content “to incite others to take desired actions” (Berger 2014, 596).
21
3) Information V alue
Information value is another factor that has been discussed. Being adapted from news
value theory, it is defined as “a property that makes news meaningful for a large audience and
that has the potential to impact others’ minds or behavior” (Rudat, Buder and Hesse 2014,
133). News value theory describes specific factors used by journalists as news selection
criteria. In Web 2.0, every Internet user has the chance to participate in news transmission
(Rudat, et al. 2014). They collect information, select it, probably edit it, and then publish the
message to their audience, simulating the process of journalism producing news. In addition,
the majority of retweeted contents are news (Kwak, Lee, Park & Moon 2010, Cited by Rudat,
et al. 2014, 133). So it is reasonable to use news value theory as a theoretical base to explain
online information transmission.
Rudat et al.’s study (2014) focused on eight news factors “that have turned out to be
stable and meaningful over time.” They found that some factors, such as relevance and
controversy, have high informational value and lead to more retweeting. They have the
potential to affect a large audience and impact the audience’s minds or behavior. In contrast,
other factors have low informational value and people are less likely to retweet such contents.
22
Table 2: List of Informational Value Factors
News Factors Meaning
1
High
Informational
Value
Unexpectedness The message is about an event or a development
that cannot be predicted or stands in contrast to
existing expectations
Relevance The message contains an event or a development
that directly affects, or will directly affect, a large
number of people
Controversy The message explicitly presents differences of
opinions
Negative
Consequences
The possible or actual negative consequences of
events are explicitly mentioned in the message
Low
Informational
Value
Aggression The message is about threatened or practiced
violence
Personalization Individuals get a special meaning within an event in
the message. One person or a few people are
illustrated or even portrayed standing for a group or
a company
Prominence The message is about a popular person, popularity
regardless of his or her actual political/economic
power
Proximity The message is about an event within a short
geographical distance
Controversy
Among these factors, controversy’s relationship to message virality is more complex.
1
Cited from Rudat, Buder & Hesse 2014, 133-134
23
Chen and Berger (2013, 584) argue that “Controversy doesn’t always increase discussion”.
Chen and Berger pointed out that a low-level controversy is likely to boost buzz in general.
But when the controversy level moves past a moderate point, additional controversy would
not trigger discussion, and might even decrease discussion. The chart below describes the
controversy – conversation relationship in a field study, which is about the online articles
from a news website (Chen and Berger 2013, 583-584). When the controversy level of the
article is relatively low (controversy score is less than 4), controversy increases the amount of
comments for the article. However, if an article is too controversial (controversy score is
more than 4), passing the moderate point, controversy would negatively influence the amount
of comments for the article.
Figure 2: Relationship between Controversy and Conversation in a Field Study
2
Controversial topics are interesting and increase people’s willingness to discuss. At the
same time, however, controversial topics can be uncomfortable to discuss. Thus “the relative
strength of these two underlying processes” (Chen and Berger 2013, 582) determines the
2
Cited from Chen & Berger 2013, 583
24
willingness to talk about controversial topics. For example, additional controversy passing
the moderate point may increase the discomfort level and decrease conversation.
4) Social Mobilization/Implication
Research also showed that messages with social implication factors are likely to
motivate people to take action. We often refer to other’s behavior to decide how to behave,
especially in an ambiguous context (Cialdini 2009, cited by Guadagno et, al. 2013, 54).
“Individuals consider an action more appropriate when they see others reacting similarly to
the situation” (Guadagno et, al. 2013, 54). In a study by Guadagno et, al. (2013), students
were presented with a blog post asking for volunteers, and they were randomly assigned to
read different comments on the blog. People who read the comments saying that other
students are willing to be volunteers are more likely to help than those who read the
comments saying that others do not want to volunteer. Another example is a large-scale
Facebook experiment (Bond et al. 2012), in which participants who were shown how many
Facebook users had voted, and pictures of their Facebook friends who had voted, were more
likely to seek voting information and vote than those participants who did not see the social
messages. This indicates that content with social implications – that is, content that tells you
what your friends are doing – may encourage people to take action.
25
2.3 Network Perspective: Connection with Others
Sharing happens in a social context. Though people are motivated by their internal status
to share, contextual factors, such as who are listening and how many people are listening,
may influence whether they would like to share and what to share. For instance, Rudat et al.
(2014, 136-137) found that people tend to tailor their retweeting information, such as
selecting specific topics, when they are aware of their audience’s interests.
Researchers believe that contextual factors play a moderate role in social transmission.
Their role is to shape when different self-serving factors and content factors matter more.
1) Anonymity
Anonymity helps to reduce people’s concern for social acceptance and feelings of threat
(Chen and Berger 2013, 587). Thus, people may react differently when their audience know,
or do not know, who they are. One example is the experiment conducted by Chen and Berger
(2013). They concluded that anonymity moderates the effect that controversy has on
conversation. When people remain anonymous, a moderate level of controversy boosts
discussion, because “it increases interest without increasing discomfort” (Chen and Berger
2013, 587). But when people disclose their identity, controversy does not increase
conversation, because people feel more uncomfortable.
26
2) Relationship Closeness
People also talk in different ways in front of close friends versus strangers. When talking
to friends, people are more likely to share their emotions, seek advice, or try to persuade
others (Berger 2014, 598-599). We tend to share more emotional experience, especially
negative emotion, with people who are close to us. Since the connection is strong, we may
feel comfortable and could expect more consolation from them. Close relationships also
facilitate social bonding (Berger 2014, 599). Since we know each other better, it is easier for
us to reinforce our opinions and reduce loneliness.
On the other hand, when talking to strangers, people tend to say more positive things
(Dubois, Bonezzi & De Angelis 2013, cited by Berger 2014, 598), or share less emotional or
controversial topics. People want to be accepted by others (Reiss 2004, cited by Berger 2014,
598). Since strangers do not know us well, we may care more about our personal image when
talking to strangers (Berger 2014, 598). Thus, we tend to talk about positive topics with them
to make ourselves look good.
3) Audience Size
Audience size is another moderator for social transmission. When talking to a small-
group audience (narrowcasting), people tend to be other-focused (Barasch and Berger 2014,
cited by Berger 2014, 599). In such situations, the audience is more “concrete and vivid”
(Barasch and Berger 2014, 287) for us. Audience’s name, story, personality, needs, etc. all
remind us to care about what others want when sharing. This may reduce the content about
27
self-presentation and emotional experience. On the other hand, however, deeper conversation
with a small group may facilitate detailed information seeking and social bonding (Berger
2014, 599).
When talking to a large group (broadcasting), the audience becomes less concrete and
people tend to be self-focused (Berger 2014, 599). In nature, “people do not consider others’
beliefs and knowledge unless something in their environment triggers them to do so” (Zhang
and Epley 2012, cited by Barasch and Berger 2014, 287). In such a situation, people may care
more about their own image and be less likely to share personal emotional experience and
reinforce shared values in front of a large-group audience (Berger 2014, 599).
4) Channel
The channel people use to communicate also influences their decisions on sharing. For
example, compared to oral communication, people are more concerned about managing their
impression and are motivated to share positive information during written communication,
which is more asynchronous and gives people some time to reflect on and revise their
messages (Berger 2014, 600)
28
III. Experiment: Test Message before the Campaign
We understood that there is a cluster of factors that may influence people to share. The
next challenge would be to learn which factors work in our own brands’ channels. The
following part of this paper will discuss the methodology helping practitioners to decide
which factors to include in order to maximize social media effectiveness.
This paper suggests that experiments can be used to test messages and provide guidance
for a social media campaign.
3.1 The Adoption of Experiments: Seldom Used in Communication Practice
The experimental method allows researchers to control the environment, variables and
subjects to establish a causal relationship between variables (Wimmer and Dominick 2003,
219). In an experiment, participants are randomly assigned to the experiment treatment
condition or the control condition (receive no treatment). All factors in both conditions are
the exact same, except for the factor(s) we want to test. This helps us to build the causal
relation so that the different outcome of each group is due to the manipulation, without any
influence caused by other variables.
In terms of the research setting, there are two types of experiment. A laboratory
experiment “is carried out on the researcher’s own turf” (Westley 1989, cited by Wimmer and
Dominick 2003, 233) and “researchers maintain tight control over the subjects’ behavior”.
29
Another type of experiment is the field experiment, which requires that the “researcher goes
to the subject’s turf” (Westley 1989, cited by Wimmer and Dominick 2003, 233) and the
setting is almost as it would be in the participants’ real life.
In general, many kinds of research have been used for communication strategic planning
to analyze situations, learn about the audience, select media channels, tailor messages, etc.
Currently, the most widely used research methods in communication practice include
secondary research, focus group meetings, in-depth interviews, observations and surveys
(Stacks and Michaelson 2010; Atkin and Freimuth, 2013).
Experiments, however, are less likely to be adopted and discussed by current
communication practitioners (Stacks and Michaelson 2010, 121; Stacks 2010, 248).
According to Stacks (2010, 248), most PR professionals would not involve themselves in
experiments. First, they lack the expertise or time to conduct a rigorous experiment. Second,
many public relations practitioners do not regard the testing method “as a major element of
their research agendas; instead, they rely on descriptive (survey and poll) and qualitative
(focus group and interview) methods to gather and analyze data” (Stacks 2010, 248).
3.2 Message Testing
Message testing, an important part of formative research, might be one of the few fields
adopting testing theory in communication practice. As Atkin and Freimuth (2001, 62)
suggested, message testing can be used to predict how effectively a message can motivate the
30
targeted audience to achieve the expected reaction. Practitioners have employed message
testing to develop communication material for decades, such as advertising content, speeches,
headlines, slogans, storyline scenarios, etc. (Atkin and Freimuth 2001, 62).
The widely used methods to test messages are in-depth interviews, focus group meetings
and surveys (Atkin and Freimuth 2001, 63). Before a campaign is launched, practitioners
might ask a sample group of people how they feel about the message through an interview,
focus group or survey.
Many campaigns also began to adopt A/B testing (also known as the split test) to test
their message effectiveness. A/B testing refers to the simple randomized experiment
involving one variable. We test the two variants (sometimes more than two variants) of this
variable, A and B, at the same time to see which one can generate better results. A successful
example is the website message test of comScore (Johnson 2014). To generate more traffic to
its product page, comScore conducted an experiment on the placement and design of
customer among 2,500 visitors. People are randomly led to the subsequent four product pages.
Results showed that Variation 1 was most attractive for viewers, increasing the conversion
rate of the page by 69% (Johnson 2014). Another good example would be the Obama email
campaign discussed at the beginning of this paper. A/B testing optimized the messages of
these campaigns and increased the effectiveness.
31
Figure 3: A/B Test of comScore
Original Product Page
Testing Product Pages
Customer Quote
32
Recently, A/B testing has been expanded into the digital field, such as web design,
digital advertisement, mobile app push notification, etc.
3.3 Social Media Message Testing: Begin to Apply Experimental Method in Social
Media
With the development of social media and the need to predict its effectiveness, some
practitioners began to test their social media message. Besides the typical research methods,
such as focus groups and interviews, for message testing, A/B testing has been applied to the
social media field.
However, as the popular social media platforms, such as Twitter and Facebook, do not
provide the function that splits the audience for a brand account, and there is no mature
third-party application to achieve this function, the existing A/B testing for social media
content is not very rigorous.
First, most current social media A/B testing collects time-series data. Some people
argued that as social media is in real-time and “only a small segment of your audience is
actually online and likely to see your tweet” (Dugan 2014) at any given time, we can publish
the testing message and control message at different times in front of different segments,
simulating the A/B testing process such that participants are assigned to the experiment group
or the control group. We can then compare data such as retweeting, favorites, comments, etc.
of these two messages to see which one generates better results. Others suggested (e.g.,
33
Zaucha 2014) that organizations can record the time they post a test message and then post
the control message at the exact same time one week later.
This method, however, ignores the possibility that the same person could be on social
media at the different times we are posting the experiment and control messages. This means
that a person probably takes part in both the test group and the control group. If this happens,
the result of the experiment will definitely be influenced. In addition, assuming that the
experiment and control messages are exposed to different groups of audience, this method
does not obey the rule that we need to randomly assign participants into groups (this principle
will be discussed in detail in chapter five. People with different social media surfing times
may feature different demographics or psychological characteristics. For example, young
people may prefer to read social media late at night, while, perhaps, middle-aged business
people tend to read social media in the morning. This means that people in the experiment
and control groups may already have different content preferences. The result of the
experiment will thus be influenced.
Second, current social media A/B testing relies on web analytics tools such as Google
Analytics to measure effectiveness, such as conversion rate from social media post to website
(Stein 2014). While these tools can help us compare the reaction of people from the
experiment and control groups, they do not calculate the statistical significance level, which
indicates the possibility that the result occurs just by chance instead of being due to our
manipulation. Statistical significance is an important indicator when analyzing the
experimental data. We could not build the causal relation between the testing message and the
34
result without statistical significance.
This paper, instead, tries to use more rigorously controlled experiments to test social
media message effectiveness in the practical context. Then, the feasibility and benefit of this
method for communication practice is discussed.
35
IV. Sample Experiments
This paper conducted three controlled experiments, to set an example about how to use
controlled experiments to test social media message effectiveness for brands.
Practical utility and emotion arousal were manipulated in the sample experiments. They
were selected from the factors introduced in Chapter Two.
Participants were recruited from Amazon Mechanical Turk, which is an online
web-based crowdsourcing marketplace coordinating human intelligence. Individuals
(providers) get paid by completing the Human Intelligence Tasks posted by requesters. Since
the Mechanical Turk allows us to recruit research subjects at a low cost, many social science
studies have begun to use it as the recruitment vehicle (Berinsky, Huber and Lenz 2012,
351-352). According to Berinsky, Huber and Lenz’s study, though Mechanical Turk providers
do not perfectly reflect the U.S. population, it can cover different types of population groups
in general. Compared to convenience sampling, which is a typical recruiting method in
experiment studies, the demographic characters of Mechanical Turk providers are more
representative and diverse. Berinsky, Huber and Lenz conclude that the Mechanical Turk
sample may not be enough to draw a conclusion for the whole population, but it is valuable
for conducting internally valid experiments.
In the experiments, participants were assigned to read tweets with different content
factors and then were asked about their attitude and reaction to the Tweets and brands. To
reduce the effect of participants’ existing bias concerning a real brand and industry in this
experiment, each participant was presented with tweets from two hypothetical brands: one is
36
Green Market, an organic food market; the other one is Pacific Airline, an airline company.
To make the brands look as real as possible, the author referred to other similar brands and
created the brand profiles.
Figure 4: Brand Profiles created for Green Market and Pacific Airline
4.1 Study One: Practical Utility and Social Media Message Effectiveness
The first sample study tests the causal impact of practical utility on social media
message effectiveness. According to the theories above, practically useful information may
trigger sharing. Therefore, the following hypotheses can be generated:
H1.1: People tend to ‘favorite’ tweets with more practical utility. Under the
#Discover section of Twitter, the tweets favorited by people we follow are
displayed. There is also a list of tweets one has favorited on his/her personal
profile. In other words, what we favorite can be seen by others on Twitter.
So this paper regards favorite as a way of sharing.
37
H1.2: People tend to share tweets with more practical utility on their own Twitter
accounts.
In addition to people’s willingness to share, this study hopes to further explore the causal
relation between practical utility and following phases of social media effectiveness, such as
influence and impact.
H1.3: People tend to follow a Twitter account with more useful tweets
H1.4: Practical utility motivates people to think positively of the brand
H1.5: Practical utility motivates people to take real purchase action, such as going
to the store, buying the product, etc.
As the theories suggested, context factors may also affect people’s decision on which
message to share. This experiment made the following hypothesis:
H1.6: Context factors moderate the impact of practical utility on Twitter message
effectiveness.
1) Method
In this experiment, two tweets were created for each brand:
Group A: Tweet with coupon information. (high practical utility)
Group B: Tweet without coupon information. (low practical utility)
All other elements of the tweets, such as wording and pictures, were the same.
Fifty people took part in this experiment. The task instruction page on Mechanical Turk
guided them to Qualtrics, where all data were collected.
38
Pretest
To collect information about context factors, participants were asked about their basic
information after landing on Qualtrics: (a) Personal information, such as gender, age; (b)
Twitter usage information, including whether they have a Twitter account (people without a
Twitter account were screened out), how familiar they are with Twitter (1 = unfamiliar, 7 =
familiar), the relationship closeness between participants and their own Twitter followers.
Experimental Treatment
Participants were then guided to read the Green Market brand profile and were randomly
assigned to the high practical utility group or the low practical utility group of this brand.
Following the posttest questions for Green Market, they were led to the Pacific Airline brand
profile and were randomly assigned to the high practical utility group or low practical utility
group of this brand.
Posttest
After participants had read the tweet of each brand, they were asked three types of
questions: (a) Willingness to engage. Participants were asked how likely they would be to
interact with the tweet, including sharing (favorite, retweet), and following the brand’s
account (1 = very unlikely, 7 = very likely); (b) Attitudes toward the brand. Participants
indicated how they thought of the brand, including the general attractiveness (1 = attractive, 5
= unattractive) and specific brand image. For Green Market, as suggested in the brand profile
and the tweets, the specific brand image is healthy (1 = healthy, 5 = unhealthy), while for
Pacific Airline, the specific brand image is with good service (1 = with good service, 5 = not
39
with good service); (c) Intention to take purchase action. Participants were asked how likely
they would be to purchase the product (1 = very unlikely, 7 = very likely).
Figure 5: Tweets Used in Study One
Green Market:
High practical utility group Low practical utility group
Pacific Airline
High practical utility group Low practical utility group
40
2) Result
T-tests showed that there is no statistically significant difference (p>0.05) between the
high- and low-practical utility groups’ results. Therefore, on Green Market and Pacific
Airline’s Twitter, practical utility may have no impact on the audience’s willingness to engage,
their attitude towards the brand and intention to take purchase action.
Table 3: Means of High- and Low- practical Utility Group: Green Market
High practical
utility group
Low practical
utility group
Sig. (p-value)
Retweet the post
(1 = very unlikely, 7 = very likely)
2.69 2.78 0.845
Favorite the post
(1 = very unlikely, 7 = very likely)
3.00 3.22 0.666
Follow Green Market’s Twitter
Account
(1 = very unlikely, 7 = very likely)
3.59 4.11 0.360
Attitude towards brand
(1=Healthy, 5=Unhealthy)
1.66 1.78 0.628
Attitude towards brand
(1=Attractive, 5=Unattractive)
1.94 2.33 0.073
Buy juice from Green Market
(1 = very unlikely, 7 = very likely)
4.38 4.78 0.389
41
Table 4: Means of High- and Low- practical Utility Group: Pacific Airline
High practical
utility group
Low practical
utility group
Sig. (p-value)
Retweet the post
(1 = very unlikely, 7 = very likely)
2.68 2.75 0.878
Favorite the post
(1 = very unlikely, 7 = very likely)
2.82 3.29 0.346
Follow Pacific Airline’s Twitter
Account
(1 = very unlikely, 7 = very likely)
3.77 3.79 0.981
Attitude towards brand
(1=With good service, 5=Not
with good service)
1.64 1.75 0.639
Attitude towards brand
(1=Attractive, 5=Unattractive)
2.05 2.11 0.777
Book Pacific Airline’s tickets (1 =
very unlikely, 7 = very likely)
4.73 4.36 0.413
4.2 Study Two: Emotion Arousal and Social Media Message Effectiveness
The second study is to test the causal relation between emotion arousal and social media
message effectiveness. This study manipulated humor. As previous studies suggested, high
emotion arousal messages—such as those that are humorous—are more likely to motivate
people to share. So this paper developed the following hypotheses:
H2.1: People tend to ‘favorite’ tweets with higher emotion
H2.2: People tend to share tweets with higher emotion arousal
As with Study One, this study planned to explore further phases of social media
effectiveness beyond willingness to share.
42
H2.3: People tend to follow a Twitter account when its posts have high emotion
arousal
H2.4: High-arousal positive emotion motivates people to think positively of the
brands. High-arousal emotions include positive ones and negative ones. As
we know, negative emotions may have little contribution to a positive brand
image. This experiment didn’t test negative emotion. So this hypothesis is
only limited to positive emotion with high arousal.
H2.5: High-arousal positive emotion motivates people to take real purchase action.
In addition, the role of context factors was also tested.
H2.6: Context factors moderate the impact of emotion arousal on social media
effectiveness.
1) Method
In this experiment, two tweets were created for each brand.
Group A: Tweet with amusement element. (high emotion arousal)
Group B: Tweet without amusement element. (low emotion arousal)
All other elements of the tweets were the same. The high-amusement elements of both
brands feature personification. In Green Market’s tweet, a personified cute orange is walking
to a blender, preparing to become juice. In Pacific Airline’s tweet, a bird is standing on the
seat of an airplane and saying he would rather take this flight than fly by himself because of
the flight’s good service. In addition, for the sake of cross-experiment study, the control
43
group tweet of the brand is same as the control group tweet in Study One.
Fifty people took part in this experiment. People who had participated in Study One
were not allowed to take part in Study Two. The Mechanical Turk instruction page guided
them to Qualtrics, where all data were collected.
Pilot Study
To test whether the tweets can achieve the intended amusement effect, a pilot study was
conducted prior to the experiment. Twenty participants (not included in the 50 people for the
main study) were randomly assigned to read the high-amusement or low-amusement tweet of
each brand. Then they rated how funny they thought the tweets were (1 = not funny at all, 7 =
very funny). Results showed that for both of the brands, the tweet from the high-amusement
group received a higher score than the tweet from the low-amusement group. This indicated
that the tweets can achieve the intended amusement effect and be used in the main
experiment.
Pretest
To collect context factors, participants were first asked about their basic information: (a)
Personal information; (b) Twitter usage information. (Same as Study One)
Experimental Treatment
They were then shown Green Market’s brand profile and were randomly assigned to the
high-amusement group or low-amusement group of this brand. Following the posttest
questions, participants were presented with Pacific Airline’s brand profile and were randomly
assigned to the high-amusement group or low-amusement group of this brand.
44
Posttest
After reading the tweet of each brand, three kinds of questions were asked: (a)
Willingness to engage; (b) Attitude towards brands; (c) Intention to take purchase action.
(Same as Study One)
Figure 6: Tweets Used in Study Two
Green Market
High emotion arousal group Low emotion arousal group
Pacific Airline
High emotion arousal group Low emotion arousal group
45
2) Result
a. Willingness to engage
For the brand of Green Market, high-arousal emotion motivates people to retweet. In
this experiment, people in the high-amusement group indicated that they would be more
likely to retweet the Green Market tweet (M=4.31) than people in the low-amusement group
(M=3.21, p<0.05).
b. Attitude towards brand
For Pacific Airline, high-arousal emotion decreases the likelihood that people consider
the brand provides good service. In this experiment, people in the high-amusement group
reported that they are less likely to consider Pacific Airline is with good service (M =2.52)
than people in the low-amusement group (M=1.91, P<0.05, 1=with good service, 5=not with
good service). This indicates that on Pacific Airline’s Twitter, high arousal emotions have
negative impact on people’s attitude towards the brand.
c. Intention to take purchase action
T-tests showed that the purchase intention difference of the high- and low- amusement
groups was not statistically significant (p>0.05). Therefore, on Green Market and Pacific
Airline’s Twitter, emotion arousal may have no impact on the audience’s intention to
purchase.
46
Table 5: Means of High- and Low- emotion Arousal Group: Green Market
High emotion
arousal group
Low emotion
arousal group
Sig. (p-value)
Retweet the post
(1 = very unlikely, 7 = very likely)
4.31 3.21 0.048
Favorite the post
(1 = very unlikely, 7 = very likely)
4.19 3.50 0.203
Follow Green Market’s Twitter
Account
(1 = very unlikely, 7 = very likely)
4.69 4.38 0.553
Attitude towards brand
(1=Healthy, 5=Unhealthy)
1.69 1.46 0.354
Attitude towards brand
(1=Attractive, 5=Unattractive)
1.73 1.71 0.918
Buy juice from Green Market
(1 = very unlikely, 7 = very likely)
4.85 4.71 0.758
Table 6: Means of High- and Low- emotion Arousal Group: Pacific Airline
High emotion
arousal group
Low emotion
arousal group
Sig. (p-value)
Retweet the post
(1 = very unlikely, 7 = very likely)
3.22 3.48 0.654
Favorite the post
(1 = very unlikely, 7 = very likely)
3.41 3.65 0.668
Follow Pacific Airline’s Twitter
Account
(1 = very unlikely, 7 = very likely)
3.26 4.04 0.180
Attitude towards brand
(1=With good service, 5=Not
with good service)
2.52 1.91 0.035
Attitude towards brand
(1=Attractive, 5=Unattractive)
2.7 2.09 0.081
Book Pacific Airline’s tickets (1 =
very unlikely, 7 = very likely)
3.78 4.22 0.376
47
4.3 Study Three: Intersection Effect of Practical Utility and Emotion Arousal on Social
Media Message Effectiveness
The third experiment tests the effect of practical utility and emotion arousal on social
media effectiveness, to provide an example of how to test more than one factor and their
interaction effect in one experiment. As previous theories suggested, practical utility and high
emotion arousal may increase sharing individually. This paper made the assumption that these
two factors together would motivate sharing and have a better effect than one factor alone.
H3.1: People tend to ‘favorite’ tweets with higher practical utility and emotion
arousal.
H3.2: People tend to share tweets with higher practical utility and emotion arousal.
Like the previous two experiments, this study planned to explore further phases of social
media effectiveness beyond willingness to share.
H3.3: People tend to follow a Twitter account when its posts have higher practical
utility and emotion arousal
H3.4: Tweets with practical utility and high emotion arousal are more likely to
motivate people to think positively of the brand
H3.5: Tweets with practical utility and high emotion arousal are more likely to
motivate people to take real purchase action.
In addition, this experiment made the assumption about the moderate role of context
factors.
H3.6: Context factors moderate the intersection impact of practical utility and high
48
emotion on social media effectiveness.
1) Method
In this experiment, four tweets were created for each brand:
Group A: Tweet with coupon information and amusement factor.
(practical utility * high emotion arousal)
Group B: Tweet with coupon information but without amusement factor.
(practical utility * low emotion arousal)
Group C: Tweet without coupon information but with amusement factor
(low-practical utility * high emotion arousal)
Group D: Tweet without coupon information and amusement factor
(low-practical utility * low emotion arousal)
All other factors of the tweets were the same. To make it easier to cross
cross-experiment study, the tweet in Group B of this experiment was the same as the tweet
from Study One Group A. The tweet in Group C of this experiment was the same as the tweet
of Group A of Study Two. The Group D tweet was the same as Group B of both Study One
and Study Two.
Fifty people took part in this experiment. People who had participated in Study One or
Two were not allowed to take part in Study Three. The Mechanical Turk instruction page
guided them to Qualtrics, where all data were collected.
49
Pretest
To collect context factors, participants were first asked about their basic information: (a)
Personal information; (b) Twitter usage information. (Same as Study One and Study Two)
Experimental Treatment
They were then shown Green Market’s brand profile and were randomly assigned to the
four groups of this brand. Following the posttest questions, participants were presented with
Pacific Airline’s brand profile and were randomly assigned to the four groups of Pacific
Airline.
Posttest
After reading the tweet of each brand, three kinds of questions were asked: (a)
Willingness to engage; (b) Attitude towards brands; (c) Intention to take purchase action.
(Same as Study One and Study Two)
2) Result
Analysis of variance (ANOV A) showed that there is no statistically significant
difference (p>0.05) among the four groups’ results. Therefore, on Green Market and Pacific
Airline’s Twitter, practical utility may have no impact on the audience’s willingness to engage,
their attitude towards the brand and intention to take purchase action.
From the experiments above, we can conclude that for Green Market, high emotion
arousal increases audience’s willingness to share. For Pacific Airline, high emotion arousal
50
negatively influence people’s attitude towards the brand. Practical utility, however, did not
impact social media effectiveness of Green Market and Pacific Airline.
As preliminary experiments, these sample experiments have their limitations. For many
dependent variables, there are difference of posttest value between the experiment group and
the control group. But the difference is not statistically significant. This is probably because
the sample size is not enough. We couldn’t tell whether the difference it due to the experiment
treatment or the sampling error. In practice, we may need a larger sample size to gain more
insights.
51
V. Practical Implications
Based on the analysis above, there are many factors that might influence people’s
willingness to share, including self-serving factors, content factors and context factors.
Communication practitioners have been trying to improve their message effectiveness using
different methods. Message testing is one of the few methods that apply testing theory in
practice. It can give more guidance on selecting and optimizing messages to achieve the
intended communication goals. With the development of social media, practitioners began to
apply message-testing methods to social media message testing, such as A/B testing.
This paper suggests that rigorously controlled experiments can also serve as an effective
method for social media message testing. The following part of this paper will discuss how to
apply the experiment method in practice.
5.1 When Can Experiments Be Used?
When practitioners have difficulties in designing messages and are not confident about
what their audience would like, experiments can help to understand which factors would be
more effective among their audience.
52
5.2 Why Are Experiments Helpful?
Compared to other research methods, experiments have following advantages:
1) Reveal the underlying factors of social media transmission.
Although practitioners have been trying to test which kinds of messages are more
effective for social media users, and many of them have had great success, few of them
understand why a specific message motivates people to share, to comment or to take action.
Their success is based on trial and error experience. As a result, the successful message
people found from the test can be only used in a single campaign. Since people do not
understand the underlying reasons that make the message successful, they are unconfident
about whether the gained success is a coincidence, or there are some factors within the
successful message which could always impact their social media effectiveness. When it
comes to a new campaign or a new phase of the same campaign, they may need to continue
testing, to find out a new message that would work for the new context.
As Amelia Showalter, the director of digital analytics of Obama campaign, pointed out,
“We were so bad at predicting what would win that it only reinforced the need to constantly
keep testing”.
Experiments, in contrast, help to establish the underlying causal relationship. Through
rigorous control over the variables and statistical significance calculation, practitioners are
confident in saying that it is the manipulation that causes the effect, without the influence of
other factors. For example, in Study Two, the author can conclude that it is the high emotion
arousal that causes the higher willingness of sharing for Green Market. Therefore,
53
experiments reveal the mechanism of why people like this message. Communication or
marketing practitioners are not only able to find out which specific message is most effective
this time, but also able to conceptually understand which kind of factors would work for our
audience in the long term.
2) Better learn about a brand’s own audience.
Not all experts’ tips or theories would be suitable for the specific social media accounts
practitioners are working for. Berger’s research (2014) and Study Two of this paper both
showed that contextual factors moderate the effect of self-serving factors and content factors
on sharing. Different brands’ accounts have different audience compositions. This may
indicate that the contextual factors, such as audience demography (the audience’s relation
closeness with their followers), of different brands are different. Thus, for practitioners, what
works for other social media accounts may not work for their brands’ social media accounts.
This is probably the reason why practical utility does not improve social media effectiveness
for Green Market, though it increases sharing in general (See Study One).
Experiments provide the opportunity to deeply learn about a brand’s own audience. In
other words, practitioners could test which factors are effective for the audience of their own
brands’ social media accounts and design more targeted, customized messages. Though other
testing methods might also achieve this function, experiments could give us more precise and
reliable information by building a causal relation.
54
3) Contribute to long-term social media strategy.
Experiments provide guidance on long-term social media strategy. By understanding the
underlying mechanism and deeply learning about the audience, practitioners can build
standards or guidelines for future social media strategy. For example, according to the
experiments in this paper, Green Market may be considered to create more high emotion
arousal tweets, because this factor can motivate its audience to retweet.
4) Could have relatively low cost.
Online tools have reduced the cost of experiments, especially in recruiting participants
and creating the experiment setting. Crowdsourcing platforms, such as Mechanical Turk,
allow practitioners to recruit a large number of participants at a low cost. For instance, Study
One of this paper recruited 50 people on Mechanical Turk, paying them $0.05 each.
Mechanical Turk charged a 10% fee. So the total cost for Study One is $2.75. In addition,
software platforms such as Qualtrics could help to create an online experiment setting.
Experiments that require little physical interaction or observation of the participants can be
conducted using this software, remotely and at low cost.

55
5.3 How to Conduct Experiments
Figure 7: Workflow of Conducting Experiments in Practical Context
Conduct the Experiment
x Present consent form
x Randomly assign
participants
x Carry out the experiment
x Collect data
Analyze and Interpret Data
x Calculate statistical significant
level
Test Hypotheses
Recruit Participants
Conduct a Pilot Study
x Check the experiment
design
Set Experiment Goals
x Related to communication
& Business goals
x What social media effects
we want to achieve
(Dependent Variables)
Develop Hypotheses
x What factors may
influence the expected
effectiveness
(Independent Variables)
x How they may
influence social media
effect (relation)
Design the Experiment
x Settings: where to conduct
x Operationalization: how to
measure variables
x Independent variables
manipulation: balance strong
manipulation & authenticity
x Experiment design:
procedure
Cluster of Theories:
Why People Share
Guidance/Pool of potential influential factors
56
1) Set experiment goals
Before starting an experiment, the first step is to identify what social media effect
expected to achieve (dependent variables) through the experiment. It might be a single phase
of the social media effectiveness, such as gaining retweets, or multiple effects, like gaining
rewteets and building a positive brand image.
Communication or business goals help to define the experiment goal. For example, if an
organization wants to promote a new product, such as Green Market in the experiments above,
the primary social media effect it might expect could be achieving a large number of retweets
and motivating people to buy. If the communication team is trying to manage a crisis and
save the brand from reputation collapse, the most important effect they want to test may be
building a positive image.
2) Develop hypotheses
The next step is to think of which factors may influence the expected effectiveness.
These factors are what will be tested in the experiment (independent variables). Is it the
emotion element of the message, a social implication factor of the tweet, or a demographic
feature and social media usage habit of our audience?
Practitioners could come up with these factors based on experience and the theories
mentioned earlier in this paper. Past social media experience may give PR or marketing
practitioners a sense of what kind of message factor probably works, or what their audience
likes. Currently, learning from previous posts might be how people select factors for
57
conducting message testing. But past experience may not be enough to generate good
hypotheses. Experience is relatively subjective and it’s hard to figure out what underlying
factors lead to the success. Thus, the factors which people feel or think that contribute to past
successful experience may not be the actual predictor of success.
This paper suggests that to select which factors to test, practitioners can refer to existing
theory and research as well as past experience. The cluster of theories regarding why people
share provides a pool of factors that will potentially increase social media effectiveness. It
could serve as a systematic and solid guidance on how to think: practitioners can examine
internal factors of the audience, content factors, and the relation between the audience and
their social media friends. By understanding the mechanisms of social transmission, they can
come up with more reliable variables that may influence social media effectiveness.
After defining the dependent variables and independent variables, practitioners can
develop a hypothesis about the relationship between these two kinds of variables, or between
selected factors and social media effectiveness. Hypotheses for experiments are usually about
causal relationships. For instance, in Study Two, one of its hypotheses was that “high emotion
arousal will increase people’s willingness to share.”
3) Design the experiment
Guided by the goal and hypothesis, practitioners can then design the experiment
procedure.
58
a. Select experiment setting
First, experiment planner need to decide where to conduct the experiment. It can be
conducted in a physical laboratory. For example, practitioners can invite people to come to
their offices and observe and test participants’ reaction for some social media contents. This
allows the experiment to remain under direct control of the researcher. But it will be costly
both in money and time.
Another option is to conduct an online experiment. Compared to experiments for other
purposes, experiments for social media require less physical treatment and observation. What
most social media experiments require, such as presenting different social media content,
asking about participants’ willingness to share, evaluating people’s attitude towards brand,
and assessing the attractiveness of the brand, can be assembled online. Using online
platforms like Qualtrics and Mechanical Turk can help to facilitate the experiment effectively
and at low cost.
b. Operationalize variables
Operationalization turns the variable into something that can be manipulated and
measured.
For independent variables, factors can be operationalized by creating specific elements
people could manipulate. For example, Study One turns the concept of practical utility into
coupon information, and manipulates this information to show, or not show, to participants.
For dependent variables, experiment designers need to operationalize the social media
59
effect expected to be achieved “by constructing scales or rules for categorizing observations
of behavior” (Wimmer & Dominick, 2003). For example, one of the dependent variables
from Study One is willingness to favorite. It is operationalized by a scale question:
Figure 8: Operationalization of People’s Willingness to Share
c. Decide how to manipulate independent variables
To manipulate independent variables, practitioners can either present the written and
verbal stimuli material to participants, or construct an event and circumstances. In the three
experiments conducted for this paper, the researcher presented written tweets to subjects.
An important principle is to make the manipulation as strong as possible and keep the
testing message as authentic as possible at the same time. To make the manipulation strong,
the difference between the different groups needs to be maximized. This helps to increase the
opportunity for independent variables to take effect. In Study Two, in order to maximize the
difference between the high-amusement group and the low-amusement group, the author
created several draft tweets to find out which was funniest. The funny tweet and not-funny
tweet were also tested in the pilot experiment to see whether the difference is obvious to other
people.
At the same time, as experiments discussed in this paper are for social media strategy in
a practical context, the testing message should appear as real as a social media message as
60
possible. This will ensure that the experiment result is valuable for the social media practice.
It might be difficult to balance authenticity and strong manipulation. For example, when
creating the testing tweet in Study Two, typical Twitter rules should be followed while trying
to make it as funny as possible, such as limiting the post to 140 words, and paying attention
to the tones, the wording, etc. In practice, the testing message should also be in accord with
the brand image and the way the brand usually talks to its publics.
Another principle is to keep all factors of different groups exactly the same, except for
the variable(s) we manipulate. This rigorous control helps to ensure that the result difference
of different groups is caused by the manipulation. For instance, as Figure 5 shows, the two
groups of tweets of Green Market were created exact the same, with the exception of the
coupon information.
d. Design the experiment
There are many types of experiment design. As an exploratory introduction for
practitioners, this paper will only discuss basic experimental design.
If practitioners only want to test one factor’s impact on social media effectiveness, the
following experiment design can be used (see Figure 9). As the diagram shows, participants
were randomly (R) assigned to two groups. Each group of people is given a pretest.
Observation or measurement results (O
1
) of the pretest are recorded. Then only the
experiment group (Group A) is exposed to the treatment or manipulation of the independent
variable (X), like reading tweet with practical utility. The control group (Group B) receives
61
no treatment. If the independent variable has two levels, like high arousal or low arousal,
each of the group receives a different treatment reflecting the different level. Then both
groups take a posttest. The difference between O
1
and O
2
of group A is compare to the
difference between O
1
and O
2
of group B.
Figure 9: Basic Experiment Design: One Independent Variable
If the pretest might influence how people answer the posttest, practitioners could
conduct the experiment without a pretest (see Figure 10). After randomly being assigned to
two groups, the two groups receive different treatments. Then both groups take the posttest
and the results are compared.
Figure 10: Basic Experiment Design: One Independent Variable without Pretest
Participants
Group A
Group B
O
1
O
1
X
1
X
2
O
2
O
2
Randomly
Assign
Same
Pretest
Different
Treatment
Same
Posttest
Participants
Group A
Group B
X
1
X
2
O
2
O
2
Randomly
Assign
Different
Treatment
Same
Posttest
62
For example, the procedure of Study One can be described as follows
3
˖ Figure 11: Basic Experiment Design: One Independent Variable-- Example
If practitioners want to involve two of more independent variables in an experiment, the
following design can be used (see Figure 12). For example, there are two variables (X, Y) to
manipulate and each of them has two levels (X
1,
X
2,
Y
1,
Y
2
). In this 2*2 experiment,
participants are randomly assigned to four groups and undergo their corresponding treatments.
The result of the posttests will be compared. This experiment design not only allows people
to explore how every independent variable influence dependent variable independently, but
also provide opportunities to investigate the interaction of the independent variables involved.
3
The pretest mentioned in the sample experiment is different from the one in the experiment design part. In the sample
experiment, pretest was to collect context information, such as gender, age, Twitter usage information. There was no
question regarding dependent variables. The pretest discussed in the experiment design part is to measure the dependent
variable before the experiment treatment and to be compared with posttest result.
Participants
High Practical
Utility Group
Read tweet with
high-practical
utility(X
1
)
Read tweet with
low-practical
utility
x Willingness to engage
x Attitude toward brand
x Intention to take action
Randomly
Assign
Different
Treatment
Same
Posttest
Low Practical
Utility Group
x Willingness to engage
x Attitude toward brand
x Intention to take action
63
Figure 12: Basic Experiment Design: Two Independent Variables
X 1 X 2
Y 1 Group A: X 1* Y 1 Group B: X 2* Y 1
Y 2 Group C: X 1* Y 2 Group D: X 1* Y 2
Study Three of this paper is an example of how to involve multiple variables in an
experiment.
Figure 13: Basic Experiment Design: Two Independent Variables--Example
High practical utility Low-practical utility
High arousal High practical utility*High arousal Low practical utility*High arousal
Low arousal High practical utility*low arousal Low practical utility*Low arousal
4) Conduct a pilot study
The initial experiment plan is not always perfect. If there is enough budget and time, it is
better to conduct a pilot study before the main experiment. Practitioners can invite a small
number of people to take part in our experiment. This will help to identify problems with the
experiment, especially with regards to whether the manipulation is strong enough and makes
sense for people.
Before the main experiment of Study Two, the author conducted a pilot experiment to
check the manipulation: whether the intended funny tweet was actually found to be funny by
people. The result showed that people believe it is funnier than the normal one. So it is safe to
64
use this tweet in the treatment group. But during this pilot study, a participant figured out that
there is a typo in the description of the Green Market brand. This showed that a pilot study
can reveal some issues of our design and increase the chances that the main experiment will
be successful.
5) Recruit participants
In the practical context, the ultimate goal to test social media message is to motivate
brands’ audience, so it is better for practitioners to recruit participants from current audience
and the potential audience of their brands. Current audience may include current social media
followers of practitioners’ organizations, current users of their products, etc. Potential
audiences are those who belong to the brand’s targeted groups, such as female college
students in Los Angeles, middle-aged male working in the technology industry. Ideally, the
participants are randomly selected and could reflect the composition of the general audience.
There are several ways to recruit participants, such as email invitation, phone invitation,
etc. The emerging crowdsourced online platforms provide an efficient, low-cost way of
recruiting. A good example is Mechanical Turk used in this paper. As we discussed before, it
is a valuable tool to conduct internally valid experiments (Berinsky, Huber and Lenz 2012).
To conduct experiments on Mechanical Turk, practitioners could create a requester account
and post the experiments. The web interface allows requester to describe the experiments to
be conducted, set requirements, and attach links of external websites, such as the Qualtrics
link used in the sample experiments. By setting specific criteria for people to take part in the
65
experiment, practitioners are able to recruit the groups we want to target.
6) Conduct the experiment
Once practitioners develop a well-defined experiment goal, polished plan, and sufficient
participants, they are ready to conduct the experiment.
At the beginning of the experiment, a consent form should be sent out to participants,
informing them about the experiment’s purpose, risks and benefits, confidentiality issues, etc.
(See the Appendix for a completed consent form used in the sample experiments). According
to the Institutional Review Board, which oversees research involving human subjects and
protects human subjects’ rights, participants have the right to refuse to take part in the
experiment. Though IRB review is only applicable for human-subjected experiments
conducted in universities, its ethical guidelines can be good rules for organizations to follow.
Only when the participant sign the consent form can experiment operators let him or her
continue the experiment.
One essential principle for experiment implementation is to randomly assign participants
to different treatment groups. Randomization helps to eliminate the extraneous factors that
might influence the result. For example, the first sample experiment in this paper opened on
Mechanical Turk at about 11:00 pm PST. At that time, while plenty of the west coast
population is still online, many on the east coast may already have gone to sleep. The first
batch of people to take part in this experiment may be located in the west. People in the east
probably see this experiment task during the next morning, when they get up. If the
66
experiment assigns the first 10 people to the treatment group and the next 10 people to the
control group, the geographic distribution of these two groups’ participants would be different,
which might influence the experiment result. Randomization could help to eliminate the
influence of geographic factors on the result.
Randomization can be achieved in several ways. Qualtrics has a built-in function of
randomization. If the experiment is conducted in a physical environment, or an online
platform without randomization function, a random number table could be used to assign
participants.
7) Analyze the data and interpret the results
After the experiment is carried out and dependent variables are measured, practitioners
can start to analyze the data. When comparing the results from different groups, it is
necessary to analyze their statistical significance level, which indicates the reliability of the
experiments’ results. P-value is the indicator of statistical significance. Only when a
statistically significant difference exists (p<0.05), can practitioners conclude that it is the
experimental treatment that causes the different results. If the difference is not statistically
significant (p>0.05), it is insufficient to build causal relation between treatment and
dependent variables. For example, in Study One, the researcher was not able to say that
practical utility motivates people to share, because the p-value is more than 0.05. In this case,
the result difference of different groups may be caused by error in sampling, or other
confounding variables.
67
If comparing the results from just two experiment groups, a t-test could be used. When
the number of experiment groups is two or more, analysis of variance (ANOV A) can be used
to analyze the statistical significance. Several kinds of statistical software, such as SPSS, and
online calculators can realize t-tests and ANOV A.
Practitioners could further explore whether there are any other factors that moderate the
effect, besides the independent variables tested initially. This can be realized through analysis
of covariance (ANCOVA).
The final step is to interpret the experiment result. As suggested before, a social media
message experiment could help practitioners to conceptually understand what their audience
likes or does not like, and which factors have an effect on the brands’ channels. This will
provide practitioners with reliable guidance and evidence to design a social media strategy.
For example, according to the three sample experiments, Green Market can learn that their
audience tends to share content with high emotional arousal, while practical utility may not
be an effective factor for motivating their audience.
68
VI. Conclusion
Practitioners have realized the importance of social media, but most of them have
difficulties in how to use social media effectively. This paper tried to address this issue in two
ways: theoretically and methodologically. Theoretically, this paper examined existing
research about social transmission, in order to help readers understand the mechanism of why
people would share in general: self-serving factors serve as internal motivations; content
factors may further trigger people’s interests; while context factors would moderate people’s
willingness and preference of sharing. On the other hand, this paper argued that controlled
experiments could serve as an effective low-cost way for practitioners to learn about their
audience and test which factors would work on their social media accounts. By conceptually
understanding the underlying factors that motivate their audience, practitioners could be
proactive and design targeted social media messages. The theories discussed at the beginning
provide guidance on how to develop experiment hypotheses, or tell people where to start to
think of the potential influential factors. To conduct an experiment, practitioners could follow
the workflow provided by this paper.
69
References
Adobe. 2013. “Digital Distress: What Keeps Marketers Up at Night?” Accessed March 20,
2015.
http://wwwimages.adobe.com/www.adobe.com/content/dam/Adobe/en/solutions/digit
al-marketing/pdfs/adobe-digital-distress-survey.pdf
ALS Association. 2014. “The ALS Association Expresses Sincere Gratitude to Over Three
Million Donors.” Accessed March 20, 2015.
http://www.alsa.org/news/media/press-releases/ice-bucket-challenge-082914.html
AMEC. 2013. “AMEC Social Media Valid Framework.” Accessed March 20, 2015.
http://amecorg.com/wp-content/uploads/2013/06/Social-Media-Valid-Framework2013
.pdf
Atkin, Charles, K. and Freimuth, Vicki. 2012. “Guidelines for Formative Evaluation Research
in Campaign Design.” In Public Communication Campaigns, edited by Rice, Ronald
E. and Atkin, Charles K, 53-68. Thousand Oaks, Calif: Sage Publications
Barasch, Alixandra and Berger, Jonah. 2014. Broadcasting and Narrowcasting: How
Audience Size Affects What People Share. Journal of Marketing Research 51, (3):
286-299.
Baumeister, Roy F., and Mark R. Leary. 1995. The Need to Belong: Desire for Interpersonal
Attachments as a Fundamental Human Motivation. Psychological Bulletin 117, (3):
497-529.
Berger, Jonah. 2011. Arousal Increases Social Transmission of Information. Psychological
Science 22, (7): 891-893.
Berger, Jonah and Katherine L. Milkman. 2012. "What Makes Online Content
Viral?" Journal of Marketing Research 49 (2): 192-205.
Berger, Jonah. 2014. Word of Mouth and Interpersonal Communication: A Review and
Directions for Future Research. Journal of Consumer Psychology 24, (4): 586-607.
Berinsky, A. J., Huber, G. A., and Lenz, G. S. (2012). Evaluating online labor markets for
experimental research: Amazon.com's mechanical turk. Political Analysis, 20(3), 351.
Boko, Gabrielle. 2014. “6 Viral-Marketing Lessons to Learn From the Ice Bucket Challenge.”
70
Entrepreneur. Accessed March 20, 2015. http://www.entrepreneur.com/article/236843
Bond, Robert M., Christopher J. Fariss, Jason J. Jones, Adam D. I. Kramer, Cameron Marlow,
Jaime E. Settle, and James H. Fowler. 2012. A 61-million-person Experiment in
Social Influence and Political Mobilization. Nature 489, (7415): 295-298.
Campo, Shelly, Natoshia M. Askelson, Erica L. Spies, Christie Boxer, Kristina M. Scharp,
and Mary E. Losch. 2013. “Wow, that was funny”: The Value of Exposure and Humor
in Fostering Campaign Message Sharing. Social Marketin Quarterly 19, (2): 84-96.
Chen, Zoey, and Jonah Berger. 2013. When, Why, and How Controversy Causes
Conversation. Journal of Consumer Research 40, (3): 580-593.
Chung, Cindy M. Y ., and Peter R. Darke. 2006. The Consumer as Advocate: Self-relevance,
Culture, and Word-of-mouth. Marketing Letters 17, (4): 269-279.
Cohen, Sheldon, and Edward Lichtenstein. 1990. Partner Behaviors that Support Quitting
Smoking. Journal of Consulting and Clinical Psychology 58, (3): 304-309.
Dichter, E. 1966. How Word-of-mouth Advertising Works. Harvard Business Review 44, (6):
147-166.
Dubois, David, Bonezzi, Andrea and De Angelis Matteo. 2013. The Dangers of Strong Ties:
How Interpersonal Closeness Influences Word-of-mouth Valence. INSEAD Working
paper.
Dugan, Lauren. 2014. “How to Use Bitly to A/B Test Your Tweets.” Accessed March 2015.
http://www.adweek.com/socialtimes/test-tweets/500664
Green, Joshua. 2012. “The Science Behind Those Obama Campaign E-Mails.” Bloomberg.
Accessed March 20, 2015.
http://www.bloomberg.com/bw/articles/2012-11-29/the-science-behind-those-obama-
campaign-e-mails
Gross, James J. 1998. "The Emerging Field of Emotion Regulation: An Integrative
Review." Review of General Psychology 2 (3): 271-299
Goodman, David J. and Preston, Jennifer. 2012. “How the Kony Video Went Viral.” The New
York Times. Accessed March 20, 2015.
http://thelede.blogs.nytimes.com/2012/03/09/how-the-kony-video-went-viral/?_r=0
Guadagno, Rosanna E., Nicole L. Muscanell, Lindsay M. Rice, and Nicole Roberts. 2013.
71
"Social Influence Online: The Impact of Social Validation and Likability on
Compliance." Psychology of Popular Media Culture 2 (1): 51-60.
Heath, Chip, Chris Bell, and Emily Sternberg. 2001. "Emotional Selection in Memes: The
Case of Urban Legends." Journal of Personality and Social Psychology 81 (6):
1028-1041.
Hennig-Thurau, Thorsten, Kevin P. Gwinner, Gianfranco Walsh, and Dwayne D. Gremler.
2004. Electronic Word-of-mouth via Consumer-opinion Platforms: What Motivates
Consumers to Articulate Themselves on the Internet? Journal of Interactive
Marketing 18, (1): 38-52.
Ives, Nat. 2012. “How Chipotle Won a Quick Lift in Twitter Mentions—Without a Single TV
Ad.” Advertising Age. Accessed March 20, 2015.
http://adage.com/article/trending-topics/chipotle-halloween-promotion-scares-twitter-
buzz/238147/
Johnson, Robin. 2014. “3 Real-life Examples of Incredibly Successful A/B Tests.” Accessed
March 2015. http://blog.hubspot.com/marketing/a-b-testing-experiments-examples
Kirchler, Erich. 1993. Spouses' Joint Purchase Decisions: Determinants of Influence Tactics
for Muddling through the Process. Journal of Economic Psychology 14, (2): 405-438.
Kwak, Haewoon, Lee, Changhyun, Park, Hosung and Moon, Sue. 2010. What is Twitter, a
Social Network or a News Media? Paper presented at the international world wide
web conference (WWW), Raleigh, NC.
Lovett, Mitchell J., Renana Peres, and Ron Shachar. 2013. On Brands and Word of Mouth.
Journal of Marketing Research 50, (4): 427-444.
Orden, Erica and Bariyo Nicholas. 2012. “How Social Media Made Joseph Kony the Web’s
Most Hated Man.” The Wall Street Journal. March 08. Accessed March 20, 2015.
http://blogs.wsj.com/speakeasy/2012/03/08/viral-video-puts-spotlight-on-uganda-rebe
l-kony/
O’Reilly, Lara. 2015. “Ntionwide Was the Most Mentioned Super Bowl Advertiser on Social
Media—But not for the Right Reasons.” Accessed March 20, 2015.
http://www.businessinsider.com/most-mentioned-super-bowl-advertisers-on-social-me
dia-data-2015-2
Petronzio, Matt. 2014. “ALS Wikipedia Page Views Up 18-Fold Since Ice Bucket Challenge.”
Mashable. September 05. Accessed March 20, 2015.
72
http://mashable.com/2014/09/06/ice-bucket-challenge-als-wikipedia/
Rainie, Lee, Hitlin, Paul, Jurkowitz, Mark, Dimock, Michael and Neidorf, Shawn. 2012. “The
Viral Kony 2020 Video.” Pew Research Center. March 15. Accessed March 20, 2015.
http://www.pewinternet.org/2012/03/15/the-viral-kony-2012-video/
Rimé, Bernard. 2007. “Interpersonal Emotion Regulation.” In Handbook of Emotion
Regulation, edited by Gross, James. J., 466–485. New York: Guilford
Rimé, Bernard. 2009. "Emotion Elicits the Social Sharing of Emotion: Theory and Empirical
Review." Emotion Review 1 (1): 60-85.
Ritson, Mark and Richard Elliot. 1999. "The Social Uses of Advertising: An Ethnographic
Study of Adolescent Advertising Audiences." Journal of Consumer Research 26 (3):
260-277.
Rudat, Anja, Jurgen Buder, and Friedrich W. Hesse. 2014. "Audience Design in Twitter:
Retweeting Behavior between Informational Value and Followers'
Interests." Computers in Human Behavior 35: 132-139.
Smith, Craig A. and Phoebe C. Ellsworth. 1985. "Patterns of Cognitive Appraisal in
Emotion." Journal of Personality and Social Psychology 48 (4): 813-838.
Stacks, Don W. 2010. “Quantitative Research Methodology: Experimental Method.” In
Primer of Public Relations Research, 248-264. New York: Guilford.
Stacks, Don W. and Michaelson, David. 2010. A Practitioner's Guide to Public Relations
Research, Measurement and Evaluation. New York: Business Expert Press.
Stein, Kiera. 2014. “How to Use Google Analytics to Run a Social Media Split Test.” Social
Media Examiner. Accessed March 20, 2015.
http://www.socialmediaexaminer.com/google-analytics-social-media-split-test/#more-
66154
Stelzner, Michael. 2014. “2014 Social Media Marketing Industry Report.” Social Media
Examiner. Accessed March 20, 2015.
http://www.socialmediaexaminer.com/social-media-marketing-industry-report-2014/
Stenovec, Timothy. 2014. “The Reasons The Ice Bucket Challenge Went Viral.” The
Huffington Post. Accessed March 20, 2015.
http://www.huffingtonpost.com/2014/08/19/ice-bucket-challenge_n_5692307.html
Suddath, Claire. 2012. “Five Reasons the Kony Video Went Viral.” Bloomberg. Accessed
73
March 20, 2015.
http://www.bloomberg.com/bw/articles/2012-03-16/five-reasons-the-kony-video-went
-viral#p2
Sundaram, DS, K. Mitra, and C. Webster. 1998. “Word-of-mouth Communications: A
Motivational Analysis.” Advances In Consumer Research VOL 25: 527-531
Tucker, Joan S. and Jennifer S. Mueller. 2000. "Spouses' Social Control of Health Behaviors:
Use and Effectiveness of Specific Strategies." Personality and Social Psychology
Bulletin 26 (9): 1120-1130.
Wimmer, Roger D and Dominick, Joseph R. 2003. Mass Media Research: An Introduction.
Belmont, Calif: Wadsworth Pub. Co.
Wojnicki, Andrea C. and Godes, David. 2008. Word-of-Mouth as Self-Enhancement. HBS
Marketing Research Paper No. 06-01.
Zaucha, Deanna. 2014. “The Fundamentals of Facebook A/B Post Testing.” Pagemodo.
Accessed March 20.
http://www.pagemodo.com/blog/fundamentals-facebook-ab-post-testing/
Zech, Emmanuelle. 1999. “Is it really helpful to verbalise one's emotions?” Gedrag en
Gezondheid 27: 42–47
74
Appendix: Consent Form Used in the Sample Experiments
Consent Form
You are invited to participate in a research study conducted by Dr. Kjerstin Thorson and Jing
Xu, MA Candidate at the University of Southern California. Your participation is voluntary.
You should read information below, and ask questions about anything you do not understand,
before deciding whether to participate. If you decide to participate, you will be asked to sign
this form.
Experiment Purpose & Procedure
This experiment is to deeply understand Twitter messages and their transmission. In the
research, you will be presented with several tweets in an online platform, Qualtrics. You will
be asked to read the tweets carefully, and then answer questions about your opinion for the
tweets. Your participation will take approximately 1.5 minute.
Risks and Benefits
There are no known risks involved in this procedure. This study will help us to better
understand communication rule of Twitter. So individuals and organizations could use Twitter
more wisely.
Payment for Participation
You will receive $0.05 for participation.
75
Confidentiality
We will keep your records for this study confidential as far as permitted by law. However, if
we are required to do so by law, we will disclose confidential information about you. The
members of the research team and the University of Southern California’s Human Subjects
Protection Program (HSPP) may access the data. The HSPP reviews and monitors research
studies to protect the rights and welfare of research subjects.
All data will be coded so that your anonymity will be protected in any research papers and
presentations that result from this work. The data will be stored at computers of research team
members, protected by personal ID and password. The data will be kept for three years.
Participation and Withdrawal
Your participation is voluntary. Your refusal to participate will involve no penalty or loss of
benefits to which you are otherwise entitled. You may withdraw your consent at any time and
discontinue participation without penalty. You are no waiving any legal claims, rights or
remedies because of your participation in this research.
Subject’s Right
If you have questions during or after the study, please contact the researcher at
xu.jngsmile@gmail.com. If you have any further questions, concerns, or complaints about
76
your rights as research participant or the research in general and are unable to contact the
research team, or if you want to talk to someone independent of the research team, please
contact the University Park Institutional Review Board (UPIRB), 3720 South Flower Street
#301, Los Angeles, CA 90089-0702, (213) 821-5272 or upirb@usc.edu
(participants will be asked to click the following button after reading the consent form)
q I have read the information provided above. I have been given a chance to ask
questions. My questions have been answered to my satisfaction, and I agree to
participate in this study.
q I disagree

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A new power: how celebrities can use social media to influence social movements

PDF

Crisis communication for tourism destinations in the new media environment

PDF

A critical assessment of the uses and effectiveness of social media in investor communications

PDF

Generating valuable content for a destination in order to reach a new generation of travelers

PDF

Developing a strategy for public relations practitioners at environmental nonprofits using insights from psychology

PDF

A study of the cultural environment of social media

PDF

Pictures on microblogs: Twitter vs. Weibo

PDF

Musicians utilizing social media to increase brand awareness, further promote their brand and establish brand equity

PDF

The impact of social media on the diabetes industry

PDF

The visual literacy explosion: a brief history, relevant cases and commonly accepted practices

PDF

Revolution for the Lulz, or, An exploration of protest in the age of new media

PDF

Strategic communications in dietary supplement marketing — a case study of the marketing of Coenzyme Q10 supplements

PDF

The share factor: implications of global digital strategy for public relations

PDF

Digital discourse in the fashion industry

PDF

Are public relations firms ready to lead in the new communication normal? the changing agency landscape and PR’s shifting roles

PDF

The new great leap forward: a two-case analysis of modern China's efforts in external communication strategies

PDF

Corporate reputation crisis in the digital age: a comparative study on Abercrombie & Fitch’s reputation crisis in the U.S., China and Taiwan

PDF

Cultural differences in communication patterns of international B2C technology companies on Weibo and Twitter: an examination of effectiveness from a motivation perspective

PDF

Roller derby, from spectacle to sport: moving past perceptions

PDF

Luring lovers: how brands make consumers fall head over heels

Asset Metadata

Creator Xu, Jing (author)

Core Title Why our audience share? Improving social media effectiveness using experiments

School Annenberg School for Communication

Degree Master of Arts

Degree Program Strategic Public Relations

Publication Date 05/11/2015

Defense Date 04/01/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag experiment,OAI-PMH Harvest,social media effectiveness

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Thorson, Kjerstin (committee chair), LeVeque, Matthew (committee member), Lynch, Brenda (committee member)

Creator Email jxu239@usc.edu,xu.jngsmile@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-568169

Unique identifier UC11300344

Identifier etd-XuJing-3444.pdf (filename),usctheses-c3-568169 (legacy record id)

Legacy Identifier etd-XuJing-3444.pdf

Dmrecord 568169

Document Type Thesis

Format application/pdf (imt)

Rights Xu, Jing

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA