Brand Crissis. Identifying Opinion Leaders in Social Media

Review of Integrative Business and Economics Research, Vol.
8, Issue 3 24
Identifying Opinion Leaders in Social Media

during Brand Crises: A Case Study on Haidilao
Hot Pot
Jilei Lin*
Wenzhou-Kean University
Yipei Huang
Jiameng Zhang
London School of Economics and Political Science
Rongjuan Chen
ABSTRACT
In social media, opinion leaders are those who can make strong influence on the public,
by posting and reposting information through online platforms. To better understand the
characteristics of online opinion leaders, we propose and test a conceptual model to
identify online opinion leaders, based on social impact theory and the theory of two-step
flow communications. Particularly, in this paper we conduct a case study on Haidilao
Hot Pot involved in a recent food-safety-related crisis. For predicting opinion leaders
during the crisis, we rely on user attributes, such as number of followers and number of
Weibo posts, as well as content-based features of 520 sample posts appeared on Sina
Weibo, which is a dominating microblogging website in China. From these posts, using
the metrics of: 1) number of reposts, 2) number of comments, and 3) number of likes,
we nominate 54 opinion leaders. In a two-step analysis, a multiple regression model
first proposes several promising factors to predict opinion leaders. Then, a logistic
regression model further confirms the characteristics of opinion leaders in this event.
Finally, we summarize our results and discuss the theoretical and practical implications
of this study.
Keywords: Opinion Leader, Brand Crisis, Online Marketing Communications, Social

Media.
Received 18 May 2018 | Revised 7 July 2018 | Accepted 2 August 2018.
1. INTRODUCTION
Social media has become an integral part of our life. Tens of millions of people use
social media every day to read breaking news, discuss social issues, and share personal
opinions. As a prominent and unique form of social media, microblogging services
enable users to create and share short messages on a real time basis. With the unique
feature of instant messaging, microblogging websites such as Twitter (twitter.com) can
generate voluminous heated communications regarding emerging society-wide topics.
As a form of news media, Twitter is probably the best-known microblogging website in
Copyright  2019 GMP Press and Printing

ISSN: 2304-1013 (Online); 2304-1269 (CDROM); 2414-6722 (Print)
Review of Integrative Business and Economics Research, Vol. 8, Issue 3 25
the world, while in China a large population use Sina Weibo (weibo.com), which is a
Twitter-like social media platform. Past research has widely recognized that social
media can facilitate online marketing communications. Following this line of research,
in this paper we aim to research the communications regarding a recent brand crisis on
Sina Weibo.
Although social media can be leveraged by marketing managers to engage consumers in

online communications, bad news widely spread on the internet can be disastrous.
During a brand crisis, bad news can trigger significant societal percussions and threaten
the public relations between a firm and its consumers. As suggested in past research,
bad news travel faster and reach wider than good news (Hornik, Satchi, Cesareo,
Pastore, 2015; Subramanian, Druschel, Chen, 1997). Thus, a better understanding of the
viral communications about a brand crisis in social media becomes essential for both
marketing practitioners and scholars.
In contrast to traditional media for marketing purpose, social media can allow marketers
and consumers to interact with each other more directly and effectively. Meanwhile,
some particular individuals play the role of opinion leader, given the wide influence
they can make on the public. According to literature, opinion leaders in general are the
important disseminators of information (Weimann, Tustin, Van Vuuren, Joubert, 2007).
In online marketing communications, opinion leaders are referred to certain individuals
and agencies who have product knowledge and serve the information needs of others
(Leonard-Barton, 1985). Past studies have claimed that online word-of-mouth
communications can be more trustworthy than brand messages (Berkman, Gilson,
1986). This is indicating that opinion leaders should play an important role in the online
communications in a brand crisis. Consequently, it is theoretically important for
marketing researchers to learn how to identify opinion leaders in social media, thereby
providing better insights into future marketing practices.
Based on our literature review, although opinion leadership has been widely discussed
in marketing literature focusing on mass media communication as well as e-WOM
(Hennig-Thurau, 2004; Richins, Root-Shaffer, 1988), little is known about how to
identify opinion leaders in social media environment, especially on a microblogging
platform such as Sina Weibo. Therefore, we are motivated to develop this research on
identifying opinion leaders in Sina Weibo, by conducting a case study of a recent brand
crisis.
Specifically, we investigate a brand crisis that suddenly emerged in August, 2017.

Using computational approaches, we explore whether user attributes (e.g., verification,
number of posts) can help us identify opinion leaders who have the power to influence
the public responses to a crisis occurred to Haidilao (haidilao.com), a famous hot pot
chain restaurant originated from mainland China. We are wishing that our research
would render theoretical contributions to the existing though limited literature on
opinion leaders in social media, but also provide practical recommendations for
marketing professionals.
2. LITERATURE VIEW AND THEORETICAL DEVELOPMENT
2.1 An Overview of Opinion Leaders
Opinion leaders are referred to individuals who have expertise in one specific area and
tend to give information and provide advice to their followers (Weimann, Tustin, Van

Vuuren, Joubert, 2007). Dating back to 1940s, scholars have already recognized the
significance of opinion leaders in forming public predilections, informing public
audience, and influencing public perceptions and behaviors (Nisbet, Kotcher, 2009).
Despite being originated from political campaigns, opinion leader is considered as a
broad concept, which has been applied in various areas where informational flow occurs
between more than two parties, such as innovation diffusion and marketing
communications (Chaney, 2001; Leonard-Barton, 1985).
Past researchers have discussed numerous interesting facts about opinion leaders.
Lazarsfeld and colleagues (1948) have depicted that opinion leaders usually track public
trends and pay attention to various affairs of their interest; also, they believe opinion
leaders are equipped with high self-efficacy in persuading others to adopt their opinions.
Other researchers have claimed that, opinion leaders are highly involved in mass media
(Myers, Robertson, 1972; Troldahl, Van Dam, 1965), highly motivated to share
information with others (Weimann, Tustin, Van Vuuren, Joubert, 2007), widely
connected with other users in social network (Park, Thelwall, 2008), and they may have
more access to ample information sources and more platforms readily accessible to
disseminate a wide range of information (Himelboim, Gleave, Smith, 2009).
In social media contexts, opinion leaders can make a strong influence on the public, by
actively participating in the flow of online information, such as posting and reposting
the latest news on a website. Particularly, opinion leaders have larger scales of social
network and higher levels of civic involvement (Rogers, 2003; Vishwanath, Barnett,
2011). With easy access to online information, nowadays ordinary users can potentially
become opinion leaders and influence mass audiences while sharing opinions with
others (Himelboim, Gleave, Smith, 2009). Past studies have suggested that, opinion
leaders can not only produce and spread information for the public, but also induce
others to disseminate the related information (Cha, Haddadi, Benevenuto, Gummadi,
2010; Phelps, Lewis, Mobilio, Perry, Raman, 2004; Weimann, Tustin, Van Vuuren,
Joubert, 2007).
2.2 Two-Step Flow of Communications
In the existing literature addressing opinion leaders, the most celebrated conceptual
model for studying opinion leadership is the classical theory of two-step flow
communications. Two-step flow theory posits that opinion leaders are capable of
influencing their close personal ties by exerting social pressure and social support
(Lazarsfeld, Berelson, Gaudet, 1948). According to literature, opinion leaders have the
four main facets as follows (Dubois, Gaffney, 2014).
1) Opinion leaders have considerable amount of followers to serve as a

receiver for opinions and information from them.
2) Opinion leaders are trusted by followers as an expert in a specific area.
3) Opinion leaders hold expertise in a specific area, which is greater or better
than average audience.
4) Opinion leaders are connected with their followers in a community.
As discussed earlier, opinion leader concept, tightly connected with word-of-mouth of

products, can be further applied in marketing practices. The rationale behind the
importance of opinion leaders in marketing communications can be explained by the
theory of two-step flow communications. For instance, brand messages sent by

marketing professionals transmit to opinion leaders who receive the information and
further forward to their followers. Given the relatively high levels of trustworthiness of
opinion leaders, they can make a difference in influencing consumer behaviors and
attitude towards the products and the brands (Buttle, 1998). In a brand crisis situation,
the role of opinion leaders can be further accentuated, as they can either reinforce or
mitigate the disastrous or favorable effects of certain online discussions related to the
crisis. Overall, the theory of two-step flow communications justifies the significance of
opinion leaders in social media, but also proposes some important attributes of opinion
leaders.
2.3 Social Impact Theory
In social media, opinion leaders are tantamount to online influentials who exert excessive
social influence on others (Baskshy, Hofman, Mason, Watts, 2011). Social impact theory,
focusing on the social influence of other parties on individuals, can well explain the
characteristics of opinion leaders.
Social impact theory indicates that for one individual, his or her feelings, attitudes and
corresponding behaviors can be positively or negatively influenced by other individuals
in the society (Latané, 1981). Social impact from actions made by certain influentials can
possibly entail alterations, in one’s behaviors in a physiological way, in one’s feelings in
an emotional and psychological way, and in one’s values in a cognitive way (Latané,
1981; Nowak, Szamrej, Latané, 1990). According to social impact theory, the magnitude
of social impact is determined by the three factors as follows.
1) Strength of the source: social presence can have different pervasive power
in terms of different authorities and positions of one individual. With
greater strength of the source, the social impact tends to be larger.
2) Social size: the size of the influential can increase the social impact one has
on others while it is not limited to physical size.
3) Immediacy: a time lag exists between the time when some information is
sent or certain actions are taken and the time when the information is
received by one individual. With a more immediate source, the social
impact tends to be larger.
The appeal of this theory arises from its generalizability in various contexts and a
testable nature to be further tested using a particular case. In a study of communication
about social events, the theory can indicate how influentials are involved in the events
and how influentials influence others over time. In particular, one’s influence on others
through social media can be specified as informational influence, which refers to the
influence realized by one’s accepting information from others (Deutsch, Gerard, 1955;
Kwak, Ge, 2012). In social media, one can search for information via keywords, gain
knowledge from posts, and make decisions regarding whether or not to accept the belief
or the knowledge of other individuals. According to social impact theory, individuals
are physiologically, psychologically, and cognitively influenced by opinion leaders. In
particular, opinion leaders can be important during a brand crisis, as consumer decisions
(e.g., purchase intention) are practically crucial to firms (Komaladewi, Indika, 2017).
2.4 Conceptual Framework

We believe that opinion leaders have certain attributes that can differentiate them from
average audiences, and those with such attributes can potentially make strong influences
Figure 1. A sample post discussing Haidilao event on Sina Weibo
on the public. Given the important role and unique features of opinion leaders, in this
study we aim to explore the attributes that identify opinion leaders in social media.
Before we start to discuss the potential attributes of opinion leaders, we should first
briefly discuss the interface and basic functions of Sina Weibo as a specific platform. In
general, Sina Weibo is a microblogging website similar to Twitter, where people read
and spread news as well as participate in wide discussion. As shown in Figure 1, a
sample post reporting the Haidilao’s hygiene and cleanliness issues has been published
on Sina Weibo on August 25, 2017.
For research purposes, we can collect message-related information as displayed,

including: 1) number of reposting, 2) number of comments, 3) number of likes, and 4)
time of posting. Also, we can retrieve user attributes from account profiles, including:
1) number of followers, 2) number of following, 3) number of Weibo posts, and 4)
verification status. Additionally, content-based features can be obtained, including: 1)
URL, 2) mention (@), and 3) hashtag (#). Then, we propose to research the behavioral
constructs of opinion leaders in a comprehensive model. In the following sub-sections,
user attributes, content-based features, as well as the indicators of the influence of
opinion leaders are discussed separately. With the proposed conceptual model, we then
conduct an empirical case study on Haidilao brand crisis.
2.4.1 Number of Followers

Number of followers, counts the number of followers that one user has. In Sina Weibo,
users can receive updated messages from those they follow. Past studies on Twitter and
Sina Weibo have noted that, number of followers is a robust measure for social
influence. Moreover, researchers have adopted number of followers as an indicator of
online opinion leader (Zhang, Zhao, Xu, 2015). Furthermore, number of followers, is
considered an appropriate measure of “social size” according to social impact theory
and “have a following” based on two-step flow hypothesis.
2.4.2 Number of Following
Number of following measures how many accounts one user follows to receive updated
information. In particular, number of following can measure the social embeddedness
addressed in two-step flow hypothesis. How many accounts one is subscribing suggests
the size of virtual community and how widely the user is connected with others. Past
research has noted that number of following can be an indicator of opinion leaders too,
but it is less robust than number of followers (Zhang, Zhao, Xu, 2015).
2.4.3 Number of Weibo Posts
Number of Weibo posts measure how many posts on Sina Weibo one user has posted
since the start of user life. Although number of posts is not directly originated from both
theories in our literature review, it is a meaningful factor because it measures social
activeness of users. Also, we consider a sub-indicator – number of Haodilao posts,
referring to how many Haidilao-related posts a user has posted during user life. The
rationale is that opinion leaders have expertise in a specific area, as suggested in
two-step flow of communications.
2.4.4 Verification Status
The capitalized “V”, a distinctive function of Sina Weibo, indicates whether one person
is verified and what status one has. Different verification status has different colors of
the letter “V” displayed after username. Verification status recognized on Sina Weibo
includes: 1) verified institution, 2) verified individual, 3) verified Daren, 4) verified
Member, and 5) non-verified. The qualification for Daren (English: Elite) and Member
is looser than the process to become verified institutions and verified individuals.
According to social impact theory, verification status of users indicates one dimension
of social impact, strength of the source. Past studies on Sina Weibo have adopted such
feature as a typical measure of social influence through social media (Nip, Fu, 2016;
Zhang, Zhao, Xu, 2015).
2.4.5 Content-based features (URL, #Hashtag, @, and Length of Weibo Posts)
First, the variable URL measures whether URL is included in a post. Typically, the
website link embedded in a Weibo post directs individuals to the website for detailed
information. Second, #Hashtag is operationalized as a dummy variable pertaining
whether any hashtag is present in the post. The variable @ is also a dummy variable
with respect to the existence of “@” to directly mention someone on Sina Weibo. Last,
length of Weibo posts is referred to the count of the characters in a post, indicating the
informativeness of the post. Although in this study we mainly focus on user attributes as

discussed earlier, content-based features have been commonly recognized in past

research, and thus, we include these factors in our exploratory analysis.
2.4.6 Time Cycle of Event
In the whole lifespan of an event, time cycle considers the exact time when a post is
published. We suppose that the full timeframe of an event can be divided into several
stages and indicate the immediacy of one’s posting and reposting related information,
according to social impact theory. In particular, individuals publishing a short message
immediately after the occurrence of an event can get considerable attention from the
public.
Based on the discussion above, Table 1 lists the promising indicators of opinion leaders
in social media during a brand crisis.
Indicators Social Impact Theory Two-Step Flow Hypothesis

Number of followers Social Size Have a following
Number of following Social embeddedness
Verification status Strength of the source
Length of Weibo posts Expertise
URL Expertise
#Hashtag Expertise
@ Social embeddedness
Time Cycle Immediacy Expertise
Number of Haidilao posts Expertise
Table 1. A list of opinion leader indicators based on past literature
2.4.7 Repost, Comment, and Like
In this study, we regard repost, comment, and like as the three metrics to differentiate
opinion leaders from ordinary users. Considering how many reposts, comments, and
likes that one individual receives from a post, the person can be defined as opinion
leader versus non-opinion-leader. These measures of opinion leader have commonly
been adopted to capture the magnitude of social impact through social media (Borge
Bravo, Esteve Del Valle, 2017).
Taken together, a conceptual framework is presented in Figure 2, in attempts to detect

opinion leaders in social media.

Figure 2. A conceptual model of identifying opinion leaders in social media
Of the factors shown in Figure 2, 1) number of followers, 2) number of following, 3)

verification status, 4) number of Haidilao posts are user attributes while 5) length of
weibo posts, 6) URL, 7) hashtag, and 8) @ are content features. In addition, 9) time
cycle of event can explain how immediately one individual responds to news and share
it with others, which is considered an attribute of user behavior on Sina Weibo.
To summarize, a comprehensive research model should include all these factors, yet we
are more focused on user attributes. Particularly, we test the hypothesis stating that the
factors as mentioned above can significantly influence the likelihood that one’s Weibo
post is reposted, commented, and liked, indicating whether an individual is opinion
leader or not.
3. METHODS
3.1 Case Description
On August 25, 2017, a news report first released by Legal Evening News appeared on
Sina Weibo, revealing some serious issues of hygiene and cleanliness related to
Haidilao, which is one of the most successful hot pot restaurants in China. In contrast to
its positive image of providing superior services to customers, the news revealed the
concealed side of a famous brand, such as Haidilao. Specifically, the news showed an
inferior level of Haidilao’s food and service quality, by displaying several pictures taken
by a reporter in its Jinsong and Taiyanggong store, including: 1) mouse appeared in the
kitchen, 2) besom washed together with tableware, and 3) soup spoons used for picking
up rubbish from the sewer. The unfavorably disgusting images shocked many citizens
and massive complaints and criticisms suddenly became viral in social media, which is
contradictory to Haidilao’s claim for the always best services to customers. These
pictures and related comments rapidly spread on Sina Weibo and were widely viewed
and forwarded by thousands of times in the following two weeks.
3.2 Data Collection
We collected information about Weibo posts and the corresponding authors, so that
user- and content-related data can be tested as the potential indicators of opinion leaders
in social media. Within a two-week time frame since the crisis suddenly emerged, 520

Weibo posts addressing Haidilao’s kitchen problems were captured by a webcrawler.

The data set contains information as follows: username, verification status of user, time
of posting, post content, number of reposts, number of comments, and number of likes
for each Weibo post captured. For each of the 520 post authors, user attributes such as
number of followees, number of followers, and number of total Weibo posts, were also
recorded in the data set. Furthermore, additional information about the 520 individuals
spreading Haidilao crisis news were retrieved through user profiles available on one’s
main page of Sina Weibo, including gender, user life, and frequency of posting. Last, as
past studies have commonly adopted content-based features such as the use of @ and #
as well as the inclusion of an URL.
3.3 Data Analysis
Our analysis aims to explore the factors that can potentially identify opinion leaders in
social media during Haidilao brand crisis. We divide the analysis into two steps: 1) use
a multiple regression model to explore these factors, and 2) develop a logistic regression
to replicate and validate the results obtained in 1).
For multiple regression modeling, we operationalize the independent variables as the

information collected from SinaWeibo by the webcrawler. Meanwhile, following past
studies we use the behavioral metrics, including number of reposts, number of
comments, and number of like, to indicate the relatively strong or weak social influence
of individuals through communications on Sina Weibo (Lee, Hosanagar, Nair, 2018;
Tafesse, 2016). Accordingly, with large number of reposts, comments, and likes, one is
considered an opinion leader while others receiving few responses are not.
In a logistic regression model, we rely on a single binary dependent variable, for each
individual to be identified as an opinion leader (= 1) or not (= 0). In contrast to least
square estimation in linear regression, logistic regression relies on maximum likelihood
estimation with error that conforms to logistic distribution. In this study, logistic
regression predicts the probability of opinion leaders, based on the following function:
where Z is a discriminant score, is the probability that a user is considered an

opinion leader, is the binary dependent variable, is the intercept of the model,
is the maximum likelihood estimator. The logistic regression further relies on a numeric
technique, Newton method, to approximate and to maximize .
For developing multiple regression and logistic regression models, the details are
presented as below (Table 2).

Dependent Variable Dependent Variable

Independent Variable
(1. Multiple Regression) (2. Logistic Regression)
1) # of followers 1) # of reposts 1) Opinion Leader
2) # of following 2) # of comments (0 = no, 1 = yes)
3) Verification status 3) # of likes
4) # of Haidilao posts
5) Length of Weibo posts
6) URL
7) Hashtag
8) @
9) Time cycle of event
Table 2. Independent variables and dependent variables used for modeling
Our results are expected to depict the main characteristics of opinion leaders, as
extracted from a multiple regression model. Then, results are further tested and
confirmed by using a logistic regression model. We suppose that a logistic regression
should yield similar results to a multiple regression, but also make our findings easy to
interpret and insightful for future research and practices.
4. RESULTS
Our results are presented as follows. First, we make the descriptive statistics of the four
user attributes proposed (i.e., # of followers, # of following, # of posts, and verification
status) with respect to the three dependent variables in question (i.e., repost, comment,
and like). Second, we report the results of two multiple linear regression models for
opinion leader identification. Third, the application of logistic regression model in
characterizing opinion leaders is discussed.
4.1 Descriptive Statistics
As social impact theory implies the importance of immediacy in determining the

magnitude of social impact, we first take a close look at the distribution of reposts,
comments, and likes, in the entire time frame of the event. Further, the importance of
time in social event progression has also motivated us to explore how crisis-related
information is spread across time. Initially, we split the time of the event, totally thirteen
days or 312 hours, on a three-hour basis. As commonly used for quantifying data of
wide range, we have taken the following adjustments on our dependent variables:
As Haidilao crisis has emerged, progressed, and deceased over time, the overall posts
discussing this event gradually reduced in size and eventually disappeared on Sina
Weibo, from our observation. Figure 3 presents three graphs realized by summing up

the amount of reposts, comments, and likes in every split time frame of the first four
days since the occurrence of Haidilao crisis on August 25, 2017.
Figure 3. The distribution of reposts, comments, and likes of Haidilao posts

The presented graphs are explicable in terms of the three major conspicuous spikes in
the whole time range. Accordingly, to capture time effects on opinion leader detection,
we have employed a new variable named “Time Cycle”, by dividing the full life span of
Haidilao crisis into four discrete time ranges: (1) Time Cycle 1: 0 to 18 hour since the
inception of the event, (2) Time Cycle 2: 18 to 51 hour, (3) Time Cycle 3: 51 to 60
hour, and (4) Time Cycle 4: 60 hour to 312 hour (i.e., until the last day that
Haidilao-related posts appeared on Sina Weibo).
Since the range of number of followers, number of following, and number of Weibo
posts is considerably large and volatile, we have adjusted these variables in the
following similar way as the method applied to number of reposts, comments, and likes:
Additionally, since repost, comment, and like are all objective measures for identifying
opinion leaders, we have combined them into one dependent variable: Opinion Leader
Index. The following formula presents how Opinion Leader Index is computed:
Since we believe a log-like relationship is more reasonable for user attributes and
continuous opinion leader index, we will adopt the metrics of: 1) followers, 2)
following, and 3) Weibos in the multiple regression to explore the factors for predicting
opinion leaders. However, to construct model in a real-world setting, we will use
unaltered number of followers, following, and Weibo posts in the logistic regression
where we use binary dependent variable rather than opinion leader index.
The following scatter plots, shown in Figure 4, present the relationships between
opinion leader index and followers, following, and Weibos, by considering verification
status. To clarify, verification status of individuals and agencies is administered by Sina
Weibo, including: 1) verified individual, 2) verified institution, 3) verified member, 4)
verified “daren” (English: Elite), and 5) non-verified. The exponential-like trend
indicates a strong correlation between opinion leader index and these three indicators.

Two-way analysis of variance (ANOVA) indicates that there are significant differences
between users with different verification status and varying followers (p < .001) and
following (p < .001); however, there is no significant difference between users with
different verification status and Weibos (p = .954).
Figure 4. The relationships between opinion leader index and user attributes
We have also noticed relatively small differences between member, daren, and
non-verified individuals in terms of opinion leader index, but they tend to be different
from verified individuals and institutions. So, we classify them into one group called
“Others”, and present the following boxplot in Figure 4 to confirm the significant
differences between verified individuals, verified institution, and other groups.
In a nutshell, the descriptive statistics presented above reveal that followers, following,
Weibos, and verification status are potentially promising factors for predicting opinion
leaders, shedding light on our further analysis.
4.2 Multiple Regression Model
4.2.1 Model 1: a comprehensive model with all proposed factors
Even if heteroskedascity interferes with accurate estimator variance, we checked that

the results of ordinary least square and generalized least square regression do not differ
in terms of statistical significance. Thus, we employed ordinary least square
no-intercept regression, which avoids dummy variable trap. According to Table 3,
followers (p < .001), following (p < .05), Weibos (p < .01), length of Weibo posts (p <

.05), number of Haidilao posts (p < .001), and time cycle (p < .01) are shown to be
insignificant. However, results suggest that verification status has no significant
relationship with opinion leader index. Other content features such as @, #hashtag, and
URL are also proved to be statistically insignificant with respect to opinion leader
index. Overall, our results present F-Statistic of 33.66, p < .001, and adjusted R-square
of .45, and turn out to be statistically robust. In addition, we adopted Akaike
information criterion (AIC) as an estimator of the relative quality of models, which
focuses on the trade-off between goodness of fit and dimensions. The results show an
AIC of 2370.00.
Variables Estimate Standard Error Significance

Daren .27 .85 t = .32, p = .75
Verified Individual -.87 .57 t = -1.53, p = .13
Verified Institution -.52 .58 t = -.89, p = .37
Verified Member -1.08 .74 t = -1.46, p = .14
Non-verified .15 .52 t = .28, p = .78
Followers 1.45 .14 t = 10.74, p < .001 ***
Following -.43 .21 t = -2.02, p < .05 *
Weibos -.52 .18 t = -2.85, p < .01 **
@ .19 .30 t = .62, p = .53
#Hashtag -.42 .22 t = -1.88, p = .06
URL .03 .35 t = .09, p = .93
Length of Weibo Posts .00 .00 t = 2.01, p < .05 *
Number of Haidilao Posts .06 .01 t = 5.83, p < .001 ***
Time Cycle -.34 .11 t = -2.99, p < .01 **
F(14,503) = 33.66, p < .001 ***, adjusted = .45, AIC = 2370.00
Note: Significance level: *** < .001, ** < .01, * < .05.
Table 3. Model 1: a comprehensive model including all factors
4.2.2 Model 2: a simplified model with selected factors
To improve model efficiency, we have attempted to eliminate those statistically

insignificant factors from Model 1 and mainly focus on user attributes along with other
related factors, such as Weibos, number of Haidilao posts, and length of Weibo posts.
Table 4 summarizes the results of the simplified model, including a short list of
potentially significant factors of our main research interest. As expected, those
statistically strong indicators of opinion leaders have remained the same in the simple
model, except length of Weibo posts cannot be considered as strongly significant (p =
.06) in Model 2. Specifically, our results show F-statistic of 43.21, p < .001, adjusted
R-square of .45, and AIC of 2368.43. Given that AIC becomes slightly lower but

adjusted R-square is not affected, we believe Model 2 is more efficient as it is simple

but remains robust.

Daren .25 .85 t = -.07, p = .94
Verified Individual -.84 .57 t = -1.91, p = .06
Verified Institution -.48 .60 t = -1,29, p = .20
Verified Member -1.04 .75 t = -1.55, p = .12
Non-verified 1.66 .53 t = .07, p = .95
Followers 1.45 .14 t = 10.79, p < .001 ***
Following -.43 .21 t = -2.24, p < .05 *
Weibos -.53 .19 t = -2.38, p < .05 *
Length of Weibo Posts .00 .00 t = 1.92, p = .06
Number of Haidilao Posts .06 .01 t = -5.88, p < .001 ***
Time Cycle -.34 .11 t = -2.72, p < .01 **
F(10,507) = 43.21, p < .001 ***, adjusted = .45, AIC = 2368.43
Table 4. Model 2: a simple model mainly focusing on user attributes
Given that verification status alone is not significantly related to opinion leader index,
we further draw a regression plot to discover the interaction effect between verification
status and followers, as the most significant factor.
As shown in Figure 5, the relationship between followers and opinion leader index
holds true only for verified individuals and verified institutions, but not for other
groups.
4.2.3 Logistic Regression
After gaining insights from previous multiple regression models, we further apply a
binomial logistic regression to identify opinion leader versus non-opinion-leader.
Although a multiple regression model can provide conclusive results regarding opinion
leader identification, in a real-world situation scholars and practitioners are more
inclined to consider a dependent variable that switches from 0 (i.e., not opinion leader)
to 1 (i.e., opinion leader). To convert the three metrics, number of reposts, number of
comments, and comment of likes, into a dummy variable, we have used the strategy as
follows: 1) determine the 7% percentile value in each metric and select observations
exceeding the value respectively, and 2) taking the three metrics together, select
observations exceeding the 7% threshold value at least in one metric and label as 1 for
opinion leader and 0 for non-opinion-leader. We end up with 54 observations standing
out as opinion leaders. The model specification is shown in the formula below:

Figure 5. Verification status moderates the relationship between

followers and opinion leader index
As shown in Table 5, we see McFadden R-square of .38 and AIC of 181.90. The results
indicate that verification status, number of followers, and number of Haidilao posts
remain to be statistically significant in differentiating opinion leaders from ordinary
users. Particularly, non-verified individuals are unlikely to be opinion leaders. For those
who have posted numerous posts related to Hadilao in the whole user life, they are more
likely to be opinion leaders. In addition, number of followers turn out to be the most
statistically robust factor of discriminating opinion leaders versus other individuals.
To evaluate the performance of our logistic regression, we have drawn receiver

operating characteristic (ROC) curve to illustrate the discriminating power of the model
applied. False positive rate measures Type I error while true positive rate indicates
predictive power. Also, we have employed area under curve (AUC), a metric of
classification accuracy, to evaluate our model. As shown in Figure 6, AUC is .9274,
which is significantly higher than the probability associated with random guess
(i.e., .50). The results indicate that opinion leader identification using the proposed
logistic regression is promising.

Daren -4.09 1.35 z = -2.64, p = .01 **

Verified Individual -.32 .72 z = .83, p = .41
Verified Institution -1.11 .80 z = -1.39, p = .16
Verified Member -.93 1.20 z = -.78, p = .44
Non-verified -3.66 1.18 z = -3.11, p < .01 **
Number of Followers .00 .00 z = 10.79, p < .001 ***
Number of Following .00 .00 z = .17, p = .87
Number of Weibo Posts -.00 .00 z = -1.14, p = .25
Number of Haidilao Posts .03 .01 z = 2.23, p = .03 *
Time Cycle .20 .24 z = .83, p = .41
McFadden = .3815, AIC = 181.90
Table 5. Logistic regression for predicting opinion leaders
Figure 6. ROC plot of logistic regression
5. GENERAL DISCUSSION
In this study, we have proposed to test the two-step flow communications and social
impact theory, by conducting a case study on Haidilao brand crisis. We have gathered

information from Sina Weibo and developed multiple regression and logistic regression
models to identify opinion leaders during the crisis. Our results present significant
indicators of opinion leader, including: number of followers, number of following,
number of Weibo posts, number of Haidilao posts, and time cycle of the event. These
findings provide evidence to accept or reject what has been suggested by past research
mainly addressing social impact theory and two-step flow communications.
Furthermore, verification status turns out to be insignificant for identifying opinion

leaders using multiple regression model; however, according to logistic regression,
verification status, at least being non-verified, is significantly related to opinion leaders.
Also, there is an interaction between number of followers and verification status in
determining opinion leaders. To extend this discussion, in the future we can directly
capture the effect of verification status on opinion leadership, as well as the interplay of
verification status and other user attributes in predicting opinion leaders. In addition,
several content-based features are found to be significantly related to opinion leader in
social media. Although in this study we are more focused on user attributes, the
characteristics of Weibo posts along with the attributes of the corresponding authors
remain to be interesting topics for future research to explore.
The contributions of this research are threefold. First, given the lack of attention to
social media communication, this study provides empirical evidence for us to better
understand brand crisis communication in social media contexts. In the academia of
brand crisis communication, a large body of studies have long been focused on
communication strategies rather than the influence of individuals (e.g., Coombs, 2006;
Coombs, Holladay, 2012; Liu, Kim, 2011). Our study has addressed this limitation.
Second, past researchers on marketing communications have overlooked the whole
event across a wide time span. In this study, we seek the opportunity to contribute to
existing literature, by studying brand crisis communications in the lifespan of the crisis
event. Third, to our best knowledge, we are the first one to combine two-step flow
hypothesis and social impact theory in a study to conceptualize opinion leaders in social
media during a brand crisis.
There are several limitations to be addressed. First, in this study we have focused on
quantitative data, such as number of followers and number of reposts, but have not yet
analyzed any qualitative data collected. A future study can rely on the available data and
conduct content analysis to gain more insightful findings, so that one can better
understand how opinion leaders influence their audiences in social media. With the
expected findings, marketers can learn how to facilitate marketing communications by
recognizing the important role of online opinion leaders. Second, for the results to be
more generalizable, a larger sample should be collected and analyzed. Particularly,
future research can explore additional cases in different settings, by considering
different cultures and different social media platforms, in attempts to replicate and
validate the current findings. Third, as marketing communications in social media are
usually two-way and involve multiple parties, including firms, general audience, and
influential individuals such as opinion leaders, a future study is suggested to examine
the influence of opinion leaders in a dynamic social media environment.
ACKNOWLEDGEMENTS
This work is supported by Wenzhou-Kean University under Grant WKU201718021.

The authors thank the faculty and students who have provided valuable comments.

REFERENCES
[1] Bakshy, E., Hofman, J. M., Mason, W. A., Watts, D. J. (2011), “Identifying
influencers on twitter”, In Proceedings of ACM International Conference on Web
Search and Data Mining (WSDM).
[2] Berkman, H., Christopher G. (1986), Consumer Behavior: Concepts and Strategies,
Boston, MA: Kent Publishing Co.
[3] Bravo, R. B., Del Valle, M. E. (2017). “Opinion leadership in parliamentary Twitter
networks: A matter of layers of interaction?”, Journal of Information Technology &
Politics, 1-14.
[4] Buttle, F. A. (1998), “Word of mouth: understanding and managing referral
marketing”, Journal of Strategic Marketing, 6(3), 241-254.
[5] Cha, M., Haddadi, H., Benevenuto, F., Gummadi, K. (2010), “Measuring user
influence in Twitter: The million follower fallacy”. In Proceedings of International
AAAI Conference on Weblogs and Social Media, George Washington University.
[6] Chaney, I. M. (2001), “Opinion leaders as a segment for marketing
communications”, Marketing Intelligence & Planning, 19(5), 302-308.
[7] Coombs, W. T. (2006), “The protective powers of crisis response strategies”,
Journal of Promotion Management, 12(3/4), 241-259.
[8] Coombs, W.T., Holladay, S. J. (2012), “Amazon-com’s Orwellian nightmare:
exploring apology in an online environment”, Journal of Communication
Management, 16(3), 280-295.
[9] Deutsch, M., Gerard, H. B. (1955), “A study of normative and informational social
influences upon individual judgment”, The Journal of Abnormal and Social
Psychology, 51(3), 629-636.
[10] Dubois, E., Gaffney, D. (2014), “The multiple facets of influence: Identifying
political influentials and opinion leaders on Twitter”, American Behavioral
Scientist, 58(10), 1260-1277.
[11] Hennig-Thurau, T., Gwinner, K. P., Walsh, G., Gremier, D. D. (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.
[12] Himelboim, I., Gleave, E., Smith, M. (2009), “Discussion catalysts in online political
discussions: Content importers and conversation starters”, Journal of Computer-
Mediated Communication, 14(4), 771-789.
[13] Hornik, J., Satchi, R. S., Cesareo, L., Pastore, A. (2015), “Information dissemination
via electronic word-of-mouth: Good news travels fast, bad news travels faster!”,
Computers in Human Behavior, 45, 273-280.
[14] Kwahk, K. Y., Ge, X. (2012), “The effects of social media on e-commerce: A
perspective of social impact theory”, In Proceedings of Hawaii International
Conference on System Sciences, 1814-1823.
[15] Komaladewi, R., Indika, D. (2017), “A Review of Consumer Purchase Decision on
Low Cost Green Car in West Java, Indonesia”, Review of Integrative Business and
Economics Research, 6(2), 172-184.
[16] Latané, B. (1981), “The psychology of social impact”, American Psychologist, 36,
343-356.
[17] Lazarsfeld, P. F., Berelson, B. R., Gaudet, H. (1948), The people’s choice: How the
voter makes up his mind in a presidential campaign, New York: Duell, Sloan &
Pierce.
[18] Lee, D., Hosanagar, K., Nair, H. S. (2018), “Advertising Content and Consumer
Engagement on Social Media: Evidence from Facebook”, Management Science (in

press).
[19] Leonard-Barton, D. (1985), “Experts as negative opinion leaders in the diffusion of
a technological innovation”, Journal of Consumer Research, 11(4), 914-926.
[20] Liu, B. F., Kim, S. (2011), “How organizations framed the 2009 HIN1 pandemic
via social and traditional media: Implications for US health communicators”,
Public Relations Review, 37(3), 233-234.
[21] Myers, J. H., Robertson, T. S. (1972), “Dimensions of opinion leadership”, Journal
of Marketing Research, 41-46.
[22] Nip, J. Y., Fu, K. W. (2016), “Challenging official propaganda? Public opinion
leaders on Sina Weibo”, The China Quarterly, 225, 122-144.
[23] Nisbet, M. C., Kotcher, J. E. (2009), “A two-step flow of influence? Opinion-leader
campaigns on climate change”, Science Communication, 30(3), 328-354.
[24] Nowak, A., Szamrej, J., Latané, B. (1990), “From private attitude to public opinion:
A dynamic theory of social impact”, Psychological Review, 97(3), 362-376.
[25] Phelps, J. E., Lewis, R., Mobilio, L., Perry, D., Raman, N. (2004), “Viral marketing
or electronic word-of-mouth advertising: Examining consumer responses and
motivations to pass along email”, Journal of Advertising Research, 44(4), 333-348.
[26] Park, H. W., Thelwall, M. (2008), “Developing network indicators for ideological
landscapes from the political blogosphere in South Korea”, Journal of Computer-
Mediated Communication, 13(4), 856-879.
[27] Richins, M. L., Root-Shaffer, T. (1988), “The role of evolvement and opinion
leadership in consumer word-of-mouth: An implicit model made explicit”, ACR
North American Advances.
[28] Rogers, E. (2003), Diffusion of Innovations (5th ed.). New York, NY: The Free
Press.
[29] Subramanian, D., Druschel, P., Chen, J. (1997), “Ants and reinforcement learning:
A case study in routing in dynamic networks”, International Joint Conference on
Artificial Intelligence.
[30] Tafesse, W. (2016), “An experiential model of consumer engagement in social
media”, Journal of Product & Brand Management, 25(5), 424-434.
[31] Troldahl, V. C., Van Dam, R. (1965), “Face-to-face communication about major
topics in the news”, Public Opinion Quarterly, 29(4), 626-634.
[32] Vishwanath, A., Barnett, A. (Eds.). (2011), The Diffusion of Innovations: A
Communication Science Perspective, New York, NY: Peter Lang.
[33] Weimann, G., Tustin, D. H., Van Vuuren, D., Joubert, J. P. R. (2007), “Looking for
opinion leaders: Traditional vs. modern measures in traditional societies”,
International Journal of Public Opinion Research, 19(2), 173-190.
[34] Zhang, L., Zhao, J., Xu, K. (2015), “Who creates trends in online social media: The
crowd or opinion leaders?”, Journal of Computer-Mediated Communication, 21(1),
1-16.


Brand Crissis. Identifying Opinion Leaders in Social Media

Uploaded by

Copyright:

Available Formats

Brand Crissis. Identifying Opinion Leaders in Social Media

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brand Crissis. Identifying Opinion Leaders in Social Media

Uploaded by

Copyright:

Available Formats

Review of Integrative Business and Economics Research, Vol.

Identifying Opinion Leaders in Social Media

Keywords: Opinion Leader, Brand Crisis, Online Marketing Communications, Social

Received 18 May 2018 | Revised 7 July 2018 | Accepted 2 August 2018.

Copyright  2019 GMP Press and Printing

Although social media can be leveraged by marketing managers to engage consumers in

Specifically, we investigate a brand crisis that suddenly emerged in August, 2017.

2. LITERATURE VIEW AND THEORETICAL DEVELOPMENT

2.1 An Overview of Opinion Leaders

Copyright  2019 GMP Press and Printing

2.2 Two-Step Flow of Communications

1) Opinion leaders have considerable amount of followers to serve as a

As discussed earlier, opinion leader concept, tightly connected with word-of-mouth of

Copyright  2019 GMP Press and Printing

2.3 Social Impact Theory

2.4 Conceptual Framework

Copyright  2019 GMP Press and Printing

Figure 1. A sample post discussing Haidilao event on Sina Weibo

For research purposes, we can collect message-related information as displayed,

2.4.1 Number of Followers

Copyright  2019 GMP Press and Printing

2.4.2 Number of Following

2.4.3 Number of Weibo Posts

2.4.4 Verification Status

2.4.5 Content-based features (URL, #Hashtag, @, and Length of Weibo Posts)

Copyright  2019 GMP Press and Printing

discussed earlier, content-based features have been commonly recognized in past

2.4.6 Time Cycle of Event

Indicators Social Impact Theory Two-Step Flow Hypothesis

Table 1. A list of opinion leader indicators based on past literature

2.4.7 Repost, Comment, and Like

Taken together, a conceptual framework is presented in Figure 2, in attempts to detect

Copyright  2019 GMP Press and Printing

Figure 2. A conceptual model of identifying opinion leaders in social media

Of the factors shown in Figure 2, 1) number of followers, 2) number of following, 3)

3.1 Case Description

3.2 Data Collection

Copyright  2019 GMP Press and Printing

Weibo posts addressing Haidilao’s kitchen problems were captured by a webcrawler.

3.3 Data Analysis

For multiple regression modeling, we operationalize the independent variables as the

where Z is a discriminant score, is the probability that a user is considered an

Copyright  2019 GMP Press and Printing

Dependent Variable Dependent Variable

Table 2. Independent variables and dependent variables used for modeling

4.1 Descriptive Statistics

As social impact theory implies the importance of immediacy in determining the

Copyright  2019 GMP Press and Printing

Figure 3. The distribution of reposts, comments, and likes of Haidilao posts

Copyright  2019 GMP Press and Printing

4.2 Multiple Regression Model

4.2.1 Model 1: a comprehensive model with all proposed factors

Even if heteroskedascity interferes with accurate estimator variance, we checked that

Copyright  2019 GMP Press and Printing

Variables Estimate Standard Error Significance

Table 3. Model 1: a comprehensive model including all factors

4.2.2 Model 2: a simplified model with selected factors

To improve model efficiency, we have attempted to eliminate those statistically

Copyright  2019 GMP Press and Printing

adjusted R-square is not affected, we believe Model 2 is more efficient as it is simple