Journal of Retailing and Consumer Services 63 (2021) 102733

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services

journal homepage: www.elsevier.com/locate/jretconser

The impact of online celebrity in livestreaming E-commerce on purchase

intention from the perspective of emotional contagion
Lu (Monroe) Meng a, 1, Shen Duan a, 1, Yijun Zhao a, *, Kevin Lü b, Siyun Chen c
a
School of Business, Renmin University of China, Beijing, China
b
Brunel University, Uxbridge, UB8 3PH, UK
c
Jinan University, Guangzhou, China

A R T I C L E I N F O A B S T R A C T

Keywords: The purpose of this study is to investigate (1) if online celebrities can directly emotionally influence consumers,
Livestream and (2) if the other audiences of the same livestream can indirectly emotionally influence consumers to increase
E-Commerce their willingness to buy the products recommended by the online celebrities. Therefore, to understand the
Emotional contagion
reasons for the successes of livestreaming e-commerce. Two studies have been conducted: in Study 1, a
Emotional trust
Admiration emotion
questionnaire-based survey has been used for data collection and AMOS version 17.0 software was used to
analyze the data of the Structural Equation Modelling (SEM); in Study 2, a web data crawler for data collection
was developed using python software, and the collected data were processed by ICTCLAS text splitting tool and
analyzed by Stata/MP 13.0 software for the econometric model constructed in this paper. This study found that
the performance of online celebrities can stimulate consumers’ emotions, and thus enhance consumers’ purchase
intention regarding the products recommended. This study also found that pleasure, arousal, admiration, and
emotional trust can influence purchase behavior for cosmetics and clothing. For food products, only pleasure
emotion, arousal emotion and admiration emotion can increase the willingness to purchase such products. For
other audiences who watch the same Online celebrity for food products, the pleasure, arousal, and admiration
emotions can increase searches for, and purchases of the products recommended. This study identified that
online celebrities’ performances can emotionally influence consumers’ willingness to purchase products rec­
ommended by online celebrities directly. In addition, the emotions of audiences watching the online celebrities
can indirectly influence consumers’ willingness to purchase products recommended by the online celebrities.

1. Introduction on the huge fan base for product promotion, it would not differ much
from the traditional marketing model. However, the product sales of
Livestreaming e-commerce has been developed rapidly (Fei et al., livestreaming promoted by Online celebrities are much higher than the
2020). Livestreaming E-Commerce is promoting and selling goods latter (Zhou et al., 2019), and the reason and mechanism behind this
through influencer streams on social media channels with charismatic, phenomenon are still unclear.
trendy online celebrities. It has elements of infomercial; variety show Only limited efforts have been made to address this issue. Existing
and group chat. The amount of goods sold through livestreaming studies on the influence of online celebrities in livestreaming e-com­
e-commerce is staggering; one Online celebrity sold 15,000 lipsticks in 5 merce on consumers’ purchasing behavior have focused mainly on the
min (Teh, 2021), and another Online celebrity has a record-high audi­ characteristics of online celebrities’ information sources (Park and Lin,
ence of more than 37 million in a month, with audiences placing orders 2020) and the performance features of online celebrities (Peng et al.,
worth millions of dollars (Yu and Hu, 2020). Many companies value the 2020), and have been restricted to the perspectives of social motivation
enormous exposure, increased volume of visits, and large audience base (Hilvert-Bruce et al., 2018) and quasi-social interaction (Sun et al.,
of online celebrities in livestreaming e-commerce for product promotion 2019).
(Chen et al., 2020). In fact, if livestream e-commerce were only relying This research considers that, unlike a traditional marketing

* Corresponding author. Business School, Renmin University of China, Room 733, Mingde Building, N0.59, Zhongguancun St. Haidian Dist., Beijing, China.
E-mail address: [email protected] (Y. Zhao).
1
Lu (Monroe) Meng and Shen Duan are co-first authors.

https://doi.org/10.1016/j.jretconser.2021.102733
Received 9 December 2020; Received in revised form 24 July 2021; Accepted 17 August 2021
Available online 30 August 2021
0969-6989/© 2021 Elsevier Ltd. All rights reserved.
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

approach, real-time interaction is one of the prominent features of audiences in real time by streaming themselves playing video games,
livestreaming e-commerce. In addition to conventional means of in­ testing food or drinks, or clothes and cosmetics and so forth in real time
teractions, such as product explanation, Question and Answer (Q&A) (Hou et al., 2019). Audiences who view the livestreams can also estab­
sessions, and lottery draws of gifts, to encourage consumers to purchase lish social communication with others by sharing and exchanging their
items, online celebrities through personality-saturated, persuasion-ori­ opinions (Kim and Kim, 2020). This shows that the main difference
ented performances generate various emotions amongst the audience, compared to traditional media is that viewing a livestream allows for
which in turn arouses, stimulates, leads, and affects viewers’ purchasing interactive participation in real time with online celebrities and other
behavior; meanwhile, other audiences’ interactions with the online ce­ audiences (Sun et al., 2019). Real-time participation satisfies the audi­
lebrities and audiences often generate echoes and have a profound in­ ence’s psychological needs for socialization and entertainment, which
fluence on the other audiences and consequently influence consumers’ leads to identification, attention, sharing, and purchasing behaviors
purchasing behavior. Thus, this study considers that the consumers’ (Park and Lin, 2020). Emotional communication refers to the expres­
purchase intentions are not merely affected by online celebrities in sion, contagion, and sharing of emotions among individuals or groups,
livestreaming e-commerce but that the livestreaming e-commerce has which starts from the transmitter’s emotions and is based on the eval­
the characteristic of diversified multi-way (amongst online celebrities uation of objective things and behavioral reactions, ultimately leading
and viewers) real-time interaction. The audiences are interested in to the emotional reactions and communication behaviors of both the
watching the livestreams and in participating in interactive communi­ transmitter and the receiver (Du et al., 2011). Environmental psychol­
cation by sending a screen bullet. The screen bullet refers to the com­ ogy considers that individuals’ behaviors interact with the environment
mentary subtitles sent and displayed simultaneously by the audience they belong to, and the virtual atmosphere of online shopping can
when watching the livestreams. This feature allows the audience to send somehow evoke consumers’ emotional reactions and ultimately deter­
real-time comments and feedback, often including emotion information, mine their behaviors (Sun et al., 2019). This study argues that the
to be seen by the online celebrities as well as all other audience mem­ communication context and interactions of livestreams and audiences
bers. As a result, it may mean that the audiences are being influenced not stimulate viewers’ emotional responses and purchasing behaviors; spe­
only by online celebrities directly but are also being influenced by the cifically, on the one hand, online celebrities influence consumer pur­
other audiences indirectly. Therefore, the research question of this study chasing behaviors by stimulating consumers’ emotional responses, and
is that considering Online celebrity and audiences of livestreaming e- on the other hand, screen bullet interactions between other audiences
commerce collectively as the subject of study, how do they emotionally and online celebrities also stimulate consumers’ emotional responses to
affect consumers’ willingness to make a purchase? change consumer purchasing behavior.
To address this question, this study applies the emotional contagion
theory to address and explain how consumers are influenced motionally 2.2. Emotional contagion
- directly and indirectly - during livestreaming e-commerce by online
celebrities and other audiences and the effect on their willingness to Emotions are essentially reactions to personal perceptions, which are
purchase the products promoted. intrinsically experiential in nature; at the same time, emotions are also a
In this research, two studies have been conducted: Study 1, through a social phenomenon that is externally manifest and can be expressed
questionnaire-based survey, examines if and how the consumers are through various forms, such as facial expressions, gestures, words, pic­
emotionally influenced by Online celebrities in livestreams, in the as­ tures, and music, and so forth (Du et al., 2011). This display of emotion
pects of pleasure emotion, admiration emotion, arousal emotion, and can be widely presented in interpersonal interactions and can have an
emotional trust, thus enhancing their purchase intention. In Study 2, impact on others, i.e., individuals can feel the emotions of others and
through collecting and analyzing real data of online product searches for react to those emotions (Fan et al., 2017). Emotional contagion is " a
and online purchases of product recommended while those products process in which a person or group influences the emotions or behavior
were being promoted in livestreams, examines how the audiences are of another person or group through the conscious or unconscious in­
influenced in terms of their purchase behavior by other audiences duction of emotion states and behavioral attitudes" (Du et al., ，2011，
through interactions including screen bullets and by Online celebrities P450). People automatically, instantly and continuously imitate others’
in terms of pleasure emotion, admiration emotion, arousal emotion, and facial expressions, voices, postures, movements, and behaviors, and so
emotional trust taking into account products in different categories. forth during social interactions and tend to capture others’ emotions at
The main contributions of this study are as follows: firstly, this study all times; the process is known as emotional contagion (Barsade et al.,
has applied and extended the theory of emotional contagion to the study 2018).
of the livestreaming e-commerce phenomenon. For consumers’ pur­ With the enrichment of emotional contagion theory, its implicational
chasing behavior, we have extended the understanding on how con­ domain has expanded from the initial field of mental health to the field
sumers are influenced in terms of their purchase intentions, where of services and marketing, and related research has focused on
pleasure emotion, admiration emotion, arousal emotion, and emotional emotional contagion between service personnel and consumers. In
trust are considered in the process of watching online celebrities. Sec­ addition, a few scholars have studied emotional contagion among con­
ond, we have taken online celebrities and audiences as a whole and have sumers. Even among consumers without direct interaction, the emotions
empirically examined the emotional effect of livestreams on consumers’ of other consumers can have an impact on consumer mood and purchase
purchase intention. intention (Chuah and Yu, 2021). Nevertheless, it is clear that most of the
We begin by introducing an overview of the studies that examine the research in the marketing field on emotional contagion has focused on
emotional impacts of livestreaming e-commerce on consumers’ pur­ offline services between "consumer-employees" and there has been less
chase decision making. Then, we develop a conceptual model of the exploration of the context of online shopping emotions. This may be
links between online celebrities, audiences, and purchase intention. because that compared to online-shopping, there are more opportunities
After presenting Study 1 and Study 2, we conclude with a discussion of for face-to-face communication and interaction between consumers and
the contributions and limitations of these studies. employees in offline shopping，so observing and measuring emotional
contagion between them are relatively straightforward. However, as
2. Theoretical background and hypotheses development many conversations and tasks increasingly take place in virtual envi­
ronments, scholars have begun to explore whether emotional contagion
2.1. Livestreaming E-Commerce and emotional communication can occur in the absence of physical location (in many cases, through
text-based interactions alone). While research in this area is still in its
Online celebrities in livestreaming e-commerce interact with infancy, existing studies suggest that emotions can actually be

2
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

transmitted virtually between individuals or teams and, more broadly, et al., 1997). The stimulation of the retail environment can stimulate
across social networking platforms (Chuah and Yu, 2021; DelVicario arousal emotions and increase consumers’ willingness to interact with
et al., 2016; Kramer et al., 2014; Magnus et al., 2021). the store environment, thus enhancing their purchase intention (Barsade
It can be seen that, along with the rise of livestreaming e-commerce, et al., 2018). This study argues that when consumers enter the state of
consumers can share information and express their emotions with online excitement due to the curiosity triggered by online celebrities’ presen­
celebrities in livestreaming broadcast platforms (Kramer et al., 2014) tation for the promoted product, the online celebrities convey arousal
and can perform a series of interactive activities, which provides the emotions to consumers (direct effect). Also, the screen bullets sent by the
necessary opportunity and condition to explore the emotional contagion audience during the livestreaming will further develop the mood of
between online celebrities and consumers in the online shopping arousal emotion (indirect effect). As a result, consumers will have
context. Therefore, this study attempts to explore the influence of online arousal emotions, which can enhance their purchase intention. Based on
celebrities on consumers’ purchasing behavior, and applies the theory of this, hypotheses H2a and H2b are proposed:
"emotional contagion" to provide understanding in this context. On the
H2a. Arousal consumer emotions, that were caused by the Online ce­
other hand, consumers may also be indirectly influenced by other au­
lebrity, will drive greater purchase intention.
diences who have been emotionally influenced by the livestreaming
broadcast through screen bullet interactions; all together, it eventually H2b. Arousal consumer emotions, that were caused by other audi­
may make consumers temporarily enter the impulsive state and increase ences, will drive greater purchase intention.
their purchase intention.
2.5. Emotional trust
2.3. Pleasure emotion
Affective trust is based on mutual interaction and attraction; it is
Consumer emotions can be categorized into several basic di­ deepened by frequent communication and exchange between in­
mensions. Pleasure, arousal, and dominance are the three bipolar di­ dividuals over time, and manifests concern for the welfare of the trusted
mensions of emotion proposed (Russell, 1980). These three dimensions person (Mian and Hattab, 2013). Emotional trust is an emotion-based
have been modified several times, and much of the literature suggests confidence formed by the trusting party with the trusted party,
that the two basic dimensions of pleasure and arousal together consti­ showing concern for the other party’s welfare and taking full account of
tute the "core emotion" (Russell et al., 1989) and that the dominance the other party’s purpose and intentions; it relies on good communica­
dimension can be removed (Donovan et al., 1994), leaving the tion (Chua et al., 2008). Consumers’ perceived trust in online retailers
remaining two dimensions as being sufficient to represent the emotional has an impact on their affective trust, which in turn, enhances purchase
response to external stimuli (Ladhari, 2007). Cheng et al. (2009) showed intentions (Zhang et al., 2014). Consumers have emotional trust in the
that pleasure and arousal were good proxies for context-induced affec­ brand community, and this increases their brand purchase intention
tive states, whereas dominance was found to have the lowest predictive (Habibi et al., 2014). With consideration of livestreaming e-commerce,
value in similar contexts. The pleasure dimension refers to the extent to this study argues that during consumers’ viewing of online celebrities’
which people feel good, happy, content, and pleased in the context performances, online celebrities will aim to gain instant favor from
(Cheng et al., 2009). When consumers feel a store environment is viewers, which leads to emotional trust (direct effect). On the other
pleasnt, they increase their willingness to purchase a product (Pihlström hand, the audience who are watching the online celebrities’ perfor­
and Brush, 2008). In this study, it is believed that, on the one hand, when mances will further strengthen the relationship with the online celeb­
online celebrities describe and present their products through friendly, rities through screen bullets shared amongst the audience, which also
warm, nice, and/or humorous language and demeanor during live­ increases interaction and participation and enhances emotional trust
streaming, the pleasure emotion will be developed amongst consumers (indirect effect). As a result, consumers will develop emotional trust,
(direct contagion). On the other hand, the audience watching live­ which will enhance consumers’ purchase intention. Based on this, hy­
streaming will respond with various positive responses through screen potheses H3a and H3b are proposed:
bullets and will also generate pleasure emotion in audience (indirect
H3a. Emotional trust, that were caused by the Online celebrity, will
contagion). Overall, pleasure emotions will be formed amongst the
drive greater purchase intention.
audience members with the online celebrities; due the enjoyable expe­
rience during livestreaming, the pleasure emotion will enhance con­ H3b. Emotional trust, that were caused by other audiences, will drive
sumers’ purchase intention. Based on this, hypotheses H1a and H1b are greater purchase intention.
proposed:
2.6. Admiration emotion
H1a. Consumer pleasure emotions, that were caused by the Online
celebrity, will drive greater purchase intention.
Admiration belongs to the category of positive psychology; it refers
H1b. Consumer pleasure emotions, that were caused by other audi­ to a high level of respect for outstanding others or role models (Becker
ences, will drive greater purchase intention. and Luthar, 2007) and is a positive emotional feeling that people
experience when they see extraordinary abilities or virtuous behavior in
2.4. Arousal emotion others (Immordino-Yang et al., 2009). Existing studies on celebrity
admiration and leadership admiration show that admiration is an
The arousal dimension refers to the extent to which people feel important factor in attracting loyal fans and mass followers. Fans’
excited, alert, stimulated, awakened, and positive in a situation (Russell, fanaticism and obsession with celebrities largely stems from their
1980). Pleasure emotions obtained by consumers in livestreaming are admiration for them (Wohlfeil and Whelan, 2012), and admiration for
referred to as the extent to which a person feels good, happy, joyful, and celebrities motivates fans to purchase products endorsed by the celeb­
satisfied in livestreaming and how therefore this enhances their product rities (Peng et al., 2020). Admiration for brands leads to more purchases,
purchasing behavior. The arousal emotion of viewers in livestreaming increased consumer attitudes and behavioral intentions, and greater
refers to the degree to which a person feels stimulated, positive, and engagement (You and Robert, 2018), connection, and loyalty (Trivedi
alert in livestreaming, which positively affects consumers’ purchase and Sama, 2020). We aim to assess to what extent consumers’ viewing of
intention (Monsuwé et al., 2004). Previous studies on store environ­ online celebrities’ performance can create emotional empathy and thus
ments have found that stores enhance consumers’ purchase intentions admiration emotion (direct effect). Also, the audience watching the
by enhancing consumers’ arousal emotions through scenography (Oliver online celebrities will express their admiration emotion for online

3
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

celebrities through screen bullets, which makes them resonate the questionnaire. From a total of 600 questionnaires, 521 question­
emotionally and further enhances the admiration emotion (indirect ef­ naires were returned. The information of the participants (survey sub­
fect). Thus, admiration emotion can enhance the psychological jects) is shown in Table 1. After excluding incomplete returned
connection with online celebrities and contribute to their purchase questionnaires, there remained 496 valid questionnaires, with an
intention for the products recommended on the site, based on which effective return rate of 82.6%. To check whether there was response bias
hypotheses H4a and H4b are proposed: in the sample, an independent sample t-test was conducted on the in­
dividual descriptive statistics of the sample, and the results showed that
H4a. Admiration emotion, that were caused by the Online celebrity,
p > 0.05, indicating that there were no significant differences in the
will drive greater purchase intention.
covariates (gender, Internet age, education, income, and job type) and
H4b. Admiration emotion, that were caused by other audiences, will that there was no response bias in the questionnaire sample. The scales
drive greater purchase intention. were all in English, which were translated into Chinese by forward and
reverse translation. (1) Forward Translation: the Chinese translation of
The framework is shown in Fig. 1.
the original English scales was conducted by two Chinese professors in
marketing, and the Chinese version was formed after repeatedly refining
2.7. Research overview the items, scoring principles, content and language. (2) Reverse Trans­
lation: 2 Chinese-American marketing experts were invited to translate
The emotional contagion theory has been applied to explain and the Chinese version back into English version, and this scale was care­
understand how consumers who watch online celebrities’ performances fully compared, reviewed, and calibrated with the original English scale,
and other audience members’ screen bullet interactions are emotionally while fine-tuning it for the Chinese national context and the Online
influenced by online celebrities themselves and other audiences. It is celebrity situation.
hypothesized that those activities and experiences would stimulate
consumers’ desire to make a purchase. In this research, two studies (2) Measurement Tools
(Study 1, and Study 2) have been conducted to examine and testify those
hypotheses. Study 1 was designed to examine how the emotional factors To ensure the reliability and validity of the questionnaire, a more
associated with livestreaming would affect consumers’ willingness to mature scale was used in this study, and modifications were made to the
purchase products (direct contagion effect), The purpose of study 1 is to scale taking into account the context of this survey. The pleasure
test hypothesis H1a-H4a; and in Study 2, since the emotional contagion emotion was referred to in the study by Cheng et al. (2009) and con­
triggered by other livestream viewers through screen bullets is difficult tained 4 question items (Cronbach’s α = 0.847). The measure of arousal
to measure in the form of self-reporting, we studied the screen bullet emotion was synthesized from Cheng et al.’s (2009) scale and included a
data of livestreaming e-commerce and examine how audiences are total of 3 items (Cronbach’s α = 0.843); the measure of emotional trust
influenced in terms of purchase intention by screen bullet interactions was synthesized from Chua et al.’s (2008) scale and included a total of 3
by other audiences along with online celebrities when different cate­ items (Cronbach’s α = 0892); the measure of admiration emotion was
gories of products (cosmetics, clothes, and food) were promoted in synthesized from Becker and Luthar (2007) scale and included a total of
livestreaming (indirect contagion effect)，The purpose of study 2 is to 3 items (Cronbach’s α = 0.866); purchase intentions were measured
test hypothesis H1b-H4b. Study 1 and Study 2 are introduced in the using a scale developed by Dodds et al. (1991) with 3 items (Cronbach’s
following section. α = 0.748). The specific scale items are shown in Table 2. All four var­
iables were measured using a 7-point Likert scale, with 1 indicating
3. Study 1 "completely disagree" and 7 indicating "completely agree". In order to
avoid the influence of other control variables, gender, Internet age,
Study 1 aimed to examine whether online celebrities’ performances education, income, and job type were selected as covariates in this
would influence consumers with the pleasure emotion, admiration study, with gender and job type treated as dummy variables and the
emotion, arousal emotion, and emotional trust, which in turn would other variables treated as continuous variables.
increase consumers’ willingness to purchase products recommended by Please refer to Table 2 for the scale question items and to Table 3 for
the online celebrities. the descriptive statistics.

3.1. Survey design

3.2. Data analysis
(1) Sample and data collection
(1) Reliability and Validity Analysis.
In this study, a survey approach was utilized. A questionnaire was In this study, AMOS 17.0 was used to test the discriminant and
distributed, and a virtual scenario was set up with reference to Park et al. convergent validity among the model variables, five of which (pleasure
(2020). The participants read the virtual scenario before filling out the emotion, arousal emotion, emotional trust, admiration emotion, and
questionnaire. In order to find suitable participants, 600 regular live­ purchase intention) were subjected to confirmatory factor analysis to
stream viewers from the offline fan community of the livestreaming e- detect the discriminant validity among the variables by comparing the
commerce platform (Douyu),2 who had previous purchase experience, strengths and weaknesses of the fit indices of the hypothetical and
were randomly recruited for this study. The text of the virtual scenario competing models. From the results of the analysis in Table 4, it was
reads: found that the fit indices of the five-factor model (χ2 = 117.437, df =
Please imagine a scenario in which you are planning to buy a pair of 109, TLI = 0.997, CFI = 0.997, RMSEA = 0.012) were significantly
sports shoes, and you have searched it for a long time in online stores but better than the other competing models. Also, the loading factors of each
have not found one you like, and at this moment, one of your favorite construct-observed variable in the model were all greater than 0.5, with
online celebrities with insightful information and personal experience of CR > 0.7 and AVE >0.5, which is shown in Tables 2 and 3.
the product happens to promote a type of sports shoes.
After reading the scenario materials, the participants then filled out 2) Test Results

A structural equation model was constructed using AMOS 17.0 to test

2
www.douyou.com. the proposed hypotheses. First, the model was tested by the chi-square/

4
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Fig. 1. Framework Chart.

Table 1
Descriptive statistics of the study sample.
Sample Characteristics Classification Criteria Sample Sample Characteristics Classification Criteria Sample

No. Percentage% No. Percentage%

Gender Male 225 45.4 Income 1-3000 Yuan 71 14.3

Female 271 54.6 3000-6000 Yuan 169 34.0
Internet age Less than 5 Years 54 11.0 6000-9000 Yuan 144 29.2
5–10 Years 66 13.3 9000+ Yuan 112 22.5
10–15Years 200 40.3 Job Type Students 168 33.9
15+ Years 176 35.4 Teachers 37 7.5
Highest Education College 89 18.0 Civil Servants 25 5.0
Undergraduate 237 47.8 Enterprises and Institutions 19 3.8
Postgraduate 93 18.7 Freelancers 247 49.8
Ph.D. 77 15.5

Table 2
Research instruments.
Measurement Measurement question items References Load Scale Reliability AVE CR
variables factor

Pleasure Emotion You feel pleased to be emotionally influenced by Online celebrities during live Cheng et al. 0.748 Cronbach’s α = 0.583 0.848
broadcast (2009) 0.847
You feel satisfied to be emotionally influenced by Online celebrities during live 0.808
broadcast
You feel relaxed to be emotionally influenced by Online celebrities during live 0.773
broadcast
You feel happiness to be emotionally influenced by Online celebrities during live 0.722
broadcast
Arousal Emotion You feel excited to be emotionally influenced by Online celebrities during live Cheng et al. 0.763 Cronbach’s α = 0.644 0.844
broadcast (2009) 0.843
You feel thrilled to be emotionally influenced by Online celebrities during live 0.804
broadcast
You feel nervous to be emotionally influenced by Online celebrities during live 0.841
broadcast
Emotional Trust You will feel that the content of live broadcast is credible when you are emotionally Chua et al., 0.728 Cronbach’s α = 0.584 0.808
affected by Online celebrities （2008） 0.807
You will feel that the products recommended by the Online celebrities are reliable 0.795
when you are emotionally affected by them during the live broadcast
You will feel that the Online celebrities are trustworthy when you are emotionally 0.768
influenced by them in the process of watching live broadcast
Admiration You will feel that the Online celebrities have excellent qualities when you are Becker and Luthar 0.751 Cronbach’s α = 0.562 0.837
emotionally influenced by them in the process of watching live broadcast （2007） 0.836
You will feel that the Online celebrities are worthy of your learning and recognition 0.754
when you are emotionally influenced by them in the process of watching live
broadcast
You will feel that you can relate to the live content of the Online celebrities when you 0.713
are emotionally influenced by them in the process of watching live broadcast
You will pay attention to the latest developments of the Online celebrities in the 0.778
process of watching them live
Purchase You are very likely to consider buying products recommended by Online celebrities Dodds （1991） 0.693 Cronbach’s α = 0.501 0.748
Intention You are willing to buy products recommended by Online celebrities 0.699 0.748
You would recommend products recommended by Online celebrities to others 0.723

2 2
degree of freedom χdf = 1.08, the superiority fit index GFI = 0.97, the sample fit indicators χdf were between 1 and 3; GFI, AGFI, NFI, TLI, and
adjusted superiority fit index AGFI = 0.963, the canonical fit index NFI CFI were greater than 0.85; and RMSEA was less than 0.08, indicating
= 0.964, the non-normal fit index TLI = 0.997, the comparative fit index that the model fit of the study was good, and further testing of the hy­
CFI = 0.997, and the root mean square error of approximation RMSEA pothesis was based mainly on each standardized path coefficient in the
= 0.012. These indicators were tested for the fit of the model, and the model.

5
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Table 3 celebrity, and thus increases purchase intention. This finding is also
Pearson correlation and the square root of ave. consistent with the findings of traditional celebrity advertising en­
Mean SD JO AR AD TR PI dorsements (Wohlfeil and Whelan, 2012; Peng et al., 2020).
The detailed product descriptions and marketing interactions of
JO 4.26 1.90 .764
AR 4.70 2.05 .102 .802 online celebrities in livestreams convey emotional trust, while eye-
AD 4.20 1.83 .154 .045 .749 catching marketing content, powerful online promotions, unique ex­
TR 4.45 1.97 .148 .122 .160 .764 planations and recommendations, and warm communication responses
PI 5.21 1.70 .375 .229 .306 .233 .705 show the progression of both parties from "meeting" to "being familiar".
Note: JO-Pleasure emotion; AR-Arouse emotions; AD- Admiration; TR- The process from "meeting" to "being familiar" is a step-by-step pro­
Emotional trust; PI- Purchase intention, The bold numbers on the diagonal are gression process that influences consumers with emotional trust, leading
the square root of AVE of the corresponding variable. to trust in the online celebrities and their recommended products, and
this then amplifies their purchase intentions. Thus, based on the
perspective of emotional contagion, the emotions of the online celeb­
Table 4 rities can be transmitted to consumers through the medium of live­
Confirmatory factor analysis results.
streams, and consumers are influenced with emotions, which influence
Model χ2 df Δχ2 TLI CFI RMSRA them to make purchase decisions.
Five factor model 117.437 109 – .997 .997 .012
(hypothesis 4. Study 2
model)
Four factor model 258.586 112 485.22** .854 .798 .061
Study 2 aims to verify that the emotions of other audiences watching
Three factor model 455.587 113 755.45** .751 .712 .077
Two factor model 855.228 114 897.25** .657 .651 .112 the Online celebrity live broadcast also influence consumers, generating
Single factor model 1113.586 115 1385.22** .554 .604 .135 pleasure emotion, arousal emotion, emotional trust and admiration,
Four factor model: the pleasure emotion and arousal emotion are combined into
which in turn increase consumers’ willingness to purchase products
one factor. Three factor model: pleasure emotion, arousal emotion and admi­ recommended by Online celebrities. In the first stage, we analyze the
ration emotion were combined into one factor. Two factor model: pleasure text of the crawled bullet screen, including the lexical classification of
emotion, arousal emotion, admiration emotion and emotional trust are com­ the bullet screen and the percentage of consumers’ attention points. In
bined into one factor. Single factor model: all variables are combined into one the second stage, we construct an econometric model to investigate the
factor. influence of other audiences’ emotions on consumers’ product search
and purchase. There are various product categories sold during the
Based on the test results of each standardized path coefficient in the Online celebrity live broadcast, but each Online celebrity tends to be
model, Table 5 shows that pleasure emotion has a significant effect on mainly good at selling one category of products. According to the Tao­
consumer purchase intention (β = 0.370, p < 0.001), arousal emotion bao Live Data Rankings released in the first quarter of 2019, the top
has a significant effect on consumer purchase intention (β = 0.201, p < three types of live broadcast goods sold were cosmetics, clothes, and
0.001), emotional trust has a significant effect on consumer purchase food. Previous studies have found that because the characteristics of the
intention (β = 0.143, p = 0.007), and admiration has a significant effect three product categories of cosmetics, clothes, and food are not consis­
on consumer purchase intention (β = 0.284, p < 0.001). Hence, hy­ tent (Moore and Konrath, 2015; Moody et al., 2010; Rodrigo et al.,
potheses H1a, H2a, H3a and H4a are validated. 2018), the degree to which they are affected by consumer sentiment can
vary. Therefore, in this study, three categories of cosmetics, clothes and
food, were selected for discussion as the research object categories. In
3.3. Results discussion each category, the three leading Online celebrities in terms of product
marketing were selected, specifically LI Jiaqi in the cosmetics category,
Therefore, this study finds that online celebrities in livestreams are Viya in the clothes category, and Little Crazy Customized House in the
able to influence audience’ emotions by purposefully planning and food category. The reason of bullet screen crawler and then make
controlling their emotional expressions, and that emotionally influenced regression is that, on the one hand, the emotional influence of other
consumers engaging in livestream interactions are able to enhance their audiences to consumers is transmitted through the form of barrage
purchase intentions. This finding also extends the traditional theory of reading by consumers; on the other hand, because the emotional influ­
emotional contagion, which was originally proposed by Hatfield et al. ence of other audiences to consumers is often just because a barrage of
(1994) to address the dynamic communication process between two two sentences causes the instantaneous state of consumers’ subcon­
face-to-face behavior entities, but the social presence generated by sciousness, it is difficult to report through consumers’ self-way of
livestreams plays an important role in the formation of behavioral measurement.
intentions.
Specifically, first, online celebrities need to not only introduce the 4.1. Text analysis
products to be sold comprehensively but also to dynamically present
marketing content and image information to audiences through product Firstly, we de-duplicate the text data and filter out the bullet screen
explanation, testing, and evaluation in attractive ways so that consumers that have no practical meaning, and filter out the bullet screen whose
are influenced by their pleasure emotions and thus draw their attention text is too short to be analyzed, such as "hmm! and "OK!" and bullet
to the products. Secondly, in livestreams, online celebrities will use in­ screen that has nothing to do with the description of products, such as
centives, such as limited-time purchases and gift distributions, which "Which song is the background music? For example, if the bullet screen
help stimulate consumers to maintain a high level of attention and in­ comments: "The lady is very beautiful, her figure is good, and her
terest in the recommended goods, with stimulated arousal emotions and makeup is exquisite", it can be divided into three phrases of "beautiful
they make purchase decisions quickly under the established time con­ person", "good figure", and "exquisite makeup". For sentences with more
straints. Thirdly, in livestreaming e-commerce, the unique features of than 15 words, a word separation tool developed by ICTCLAS was used
online celebrities, such as appearance, language, personality, talent, and for the operation. Secondly, by compiling a marketing lexicon of live
preference of netizens are regarded as opinion leaders and idols by some broadcast emotion characteristics and product characteristics, the
consumers. These features will inspire consumers’ admiration, which is number of mentions and percentage statistics of the bullet screen were
transmitted to the recognition of the product advertised by the Online calculated. Finally, the words that have been sub-worded are further

6
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Table 5
Model path coefficients.
Pathway Relation Non-standardized coefficient Standardization coefficient P-value Result

H1a: Pleasure emotion→purchase intention + 0.309 0.370 <0.001 Support

H2a: Arousal emotion → purchase intention + 0.160 0.201 <0.001 Support
H3a: Emotional trust → purchase intention + 0.123 0.143 0.007 Support
H4a: Admiration → purchase intention + 0.245 0.284 <0.001 Support

compared to the marketing lexicon, categorized into the corresponding acquisition; secondly, the crawling time is in the first quarter of 2019. In
marketing lexicon’s characteristic categories, and the final word number order to control the consistency of the collection time, we selected the
and percentage statistics are counted. time point when the target Online celebrities were conducting live
broadcast at the same time as the time node for bullet screen data
4.2. Marketing lexicon compilation collection, and the final number of days for bullet screen data collection
was 288 days, with a total of 1456,372 useful bullet screens crawled. Li
The existing emotion lexicon construction considers emotional po­ Jiaqi, an Online celebrity in the cosmetics category, crawled 513,543
larity and intensity, but lacks research on emotional marketing effects. bullet screens in total. The clothes category live broadcast Online ce­
In this paper, the construction of the emotion marketing effect lexicon is lebrity is Viya, with a total of 499,289 bullet screens. The food category
completed by the following process, as shown in Fig. 2 live broadcast Online celebrity is Little Crazy Customized House, with
In the first step, an expert method was used to label the emotion 443,540 bullet screens.
marketing effect on the basis of the Dalian University of Technology
Ontology Lexicon of Emotion Words, which contains 7 major categories 4.4. Data analysis
and 21 subcategories, with a total of 27,467 words and phrases labeled.
In the second step, a baseline emotion marketing effect lexicon was The data analysis also contains two parts: the first part being the
formed by synthesizing the Chinese positive and negative word usage lexical classification and statistical processing of the overall crawled
lexicon, the CNKI polarity emotion word lexicon, and the National bullet screens; the second part being the analysis of the empirical results
Taiwan University Sentiment Dictionary (NTUSD), as well as using a of the econometric model.
word forest of synonyms to expand the text emotion words. Finally, on
the basis of the baseline emotional marketing effect words, the unknown (1) Text Analysis Results
words were classified by using Pointwise Mutual Information (PMI) to
crawl to the live broadcast bullet screen, thus making the emotional The percentage analysis of 1456,372 useful bullet screen comments
lexicon further extended. In the semantically clear words were manually shows that 99,033 (68.0%) of the consumer comments involve emo­
selected as the baseline emotional marketing effect words. Then, equa­ tions, which far exceeds the 466,039 comments (32.0%) on product
tions (1) and (2) were used to calculate and categorize them. features. In terms of emotions, consumers in the comments focused on
∑
m
p(w,xi )
the admiration towards the live broadcast Online celebrities (31.30%).
log 2 p(w)p(x i) In terms of product features, consumers in comments focused on the
S(w) = i=1
(1) perceived value of the products (48.50%). By looking at the total per­
m
centage of the number of consumer concerns, it shows that consumers’
∑
m
p(w,yi )
log 2 p(w)p(y concern about the products recommended by live broadcast Online ce­
i)
lebrities depends more on emotions, and the percentage distribution of
S(w) = i=1 (2)
m the consumer concerns is shown in Table 7.
In the sentiment word categorization formula S(w), m and n repre­ The text analysis verified that consumers’ attention to Online ce­
sent the number of positive and negative baseline words, respectively, p lebrities’ products mainly depends on emotional nature, and product
(w) is the probability size of the occurrence of unidentified words, p(xi) characteristics play a secondary role in this process; therefore, in the
and p(yi) represent the probability size of the occurrence of positive and econometric model, the emotion of the audience watching Online ce­
negative baseline words, respectively, p(w, xi) represents the probability lebrity live broadcast is studied as the independent variable in this study,
distribution of the co-occurrence of unidentified words as well as posi­ while product characteristics are used as control variables for model
tive baseline words, and p(w, yi) is the probability distribution of the co- construction.
occurrence of unidentified words with negative words. The final ob­
tained marketing lexicon is shown in Table 6 below. 4.5. Econometric model construction & variable definition

4.3. Data source The emotions of other audiences watching Online celebrity live
broadcasts also influence consumers, generating pleasure emotions,
The data used in this study are all objective data, and the source of arousal emotions, emotional trust and admiration, which in turn
the data obtained is the real-time search volume and sales volume of the enhance consumers’ willingness to purchase products recommended by
Taobao online stores of the live broadcast Online celebrities selected for online celebrities. Therefore, in the modeling process, the emotion
this study for a period of one year from April 2019 to April 2020 in variables of other audiences watching Online celebrity live broadcast,
cooperation with Zhi Gua Data.3 the number of consumer searches and purchase volume need to be
In this study, three categories of products, being cosmetics, clothes defined.
and food, were selected as the research objects according to the study.
The specific process is as follows: in order to avoid the interference (1) Emotions of Audience Watching Online Celebrity Live Broadcast
triggered by different platforms of the same category of products, we
selected each category from the same live broadcast platform for The Online celebrity representatives of three product types (cos­
metics, clothes and food) were selected as the research objects, and the
PYTHON data crawler was used to capture the Online celebrity live
3
https://www.zhigua.cn/. bullet screen from each Online celebrity’s corresponding live broadcast

7
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Fig. 2. Words-Segmentation flow chart.

the difference was that the process analyzed the bullet screen crawled by
Table 6
the three types of Online celebrities separately). The four sub-
Marketing word list.
dimensions of this study are pleasure emotion (JO), arousal emotion
Marketing Lexicon Words that correspond to the corresponding dimensions of (AR), emotional trust (TR) and admiration (AD).
Dimensions the marketing lexicon

JO Delighted, cheerful, joy, smiling, hi, beautiful, happy, (2) Number of Searches and Purchases of Products
pleased … …
AR Flustered, panic, overwhelmed, frazzled, miracle, wow,
goddammit, eye-opening … … In this study, the search and purchase volumes about the recom­
TR useful, good, bought, paragraph, buy, screenshot, cheat, mended products were provided by Zhi Gua data, the data screening
counter, genuine, really, sure, trust, convinced, believe, start node is when the recommended products were launched after the
trustworthy, reliable … … start of live broadcast session, and the end node is the end time of the
AD Good-looking, praise, beautiful, face, wow, nice,
handsome, cute, lady, fan, good show, awesome,
live broadcast. The search and purchase volumes of the recommended
invincible, good skills, good player 6 6666 … … products during this period were observed, which were used as the
Perceived Quality Good quality, very soft, rough, nice … … search and purchase data of the recommended products by the audience
Perceived Value Good value, high price, expensive … … watching the live broadcast. In addition, through their own demon­
Product Style Size, color, big … …
stration and introduction during the live broadcast, the Online celeb­
rities of the three selected products hope that the consumers watching
platform website. To ensure the uniformity of data search time, the the live broadcast will enter their own online stores or agent online
crawling time was chosen to capture the live broadcast bullet screen stores to purchase the corresponding introduced products. Therefore,
during the same live broadcast time for each Netflix. After that, the live this study starts from the beginning of bullet screen crawling and focuses
bullet screen of three different product types were analyzed separately. on the number of searches for products by consumers watching the live
The words corresponding to different bullet screen statements in the text broadcast. To ensure that the number of searches and purchases are
were divided into the corresponding emotional dimensions, and the indeed triggered by consumers watching the live broadcast, we match
daily total number of words in each sub-dimension was counted (the the bullet screen crawling time with the search purchase data provided
process was the same as the text analysis process in the first stage, and by Zhi Gua. The number of searches for the product is defined as SN.

Table 7
Percentage distribution of consumer concerns.
Online celebrities Emotional Characteristics （990333，68%） Online celebrities Product Characteristics （466039，32%）

Indicators JO AR AD TR Perceived Quality Perceived Value Product Style

No. of being mentioned 181528 281255 309974 217576 117442 226029 122568
Percentage of being mentioned 18.33% 28.41% 31.30% 21.96% 25.20% 48.50% 26.30%

8
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Similarly, tracking the search page yields whether the consumer even­
∑
2 ∑
2 ∑
2
tually purchased the product, and thus the purchase volume of the PN = C0 + C1 + C2 + σ21j JO • Ij + σ 22j AR • Ij + σ 23j AD • Ij
product is counted, defined as PN. j=0 j=0 j=0
This study uses the dummy variable IJ to indicate the product type, ∑
2
which is used to distinguish the influence of different types of emotions + σ 24j TR • Ij + +λ21 PP + ε2 (4)
on consumers’ product search behavior, J = 0,1,2, I0 for cosmetics j=0

category live broadcast Online celebrity, I1 for clothes category live

where, C0, C1 and C2 are intercept terms; When the value of j is 0, σ110 ,
broadcast Online celebrity, and I2 for food category live broadcast On­
σ 120 , σ130 , σ 140 are the linear influences of pleasure emotion, arousal
line celebrity. When I0 = 1, I1 = 0, I2 = 0, the dependent variables are
emotion, admiration and emotional trust on the search of products by
the search behavior and purchase behavior of cosmetics class live
the audience watching cosmetics category live broadcast, respectively;
broadcast Online celebrity; I0 = 0, I1 = 1, I2 = 0, the dependent vari­
σ 210 , σ220 , σ 230 , σ240 are the linear influences of pleasure emotion,
ables are the search behavior and purchase behavior of clothes class live
arousal emotion, admiration and emotional trust on the purchase of
broadcast Online celebrity; I0 = 0, I1 = 0, I2 = 1, the dependent vari­
products by the audience watching cosmetics category live broadcast,
ables are the search behavior and purchase behavior of food class live
respectively. When the value of j is 1, σ111 , σ 121 , σ 131 , σ141 are the linear
broadcast Online celebrity. There is no case that I0, I1, and I2 are 1 at the
effects of pleasure emotion, arousal emotion, admiration and emotional
same time or two of them are 1.
trust on the search of products by the audience watching clothes cate­
According to the lexical analysis of the crawled bullet screen, the
gory live broadcast, and σ211 , σ221 , σ 231 , σ 241 are the linear effects of
filtered bullet screen includes two major parts: the emotional charac­
pleasure emotion, arousal emotion, admiration and emotional trust on
teristics of the live broadcast audience and product characteristics. The
the purchase of products by the audience watching clothes category live
lexical ratio analysis reveals that compared with the emotional charac­
broadcast, respectively. When the value of j is 2, σ112 , σ 122 , σ 132 , σ 142 are
teristics, the product characteristics are not the primary reason for
the linear effects of pleasure emotion, arousal emotion, admiration and
consumers to pay attention to the products, but only play a secondary
emotional trust on the search of products by the audience watching the
role in the influence process. Therefore, product characteristics are
food category live broadcast, respectively; σ212 , σ 222 , σ232 , σ242 are the
considered as control variables in this study. According to the bullet
linear effects of pleasure emotion, arousal emotion, admiration and
screen subscripts and marketing lexicon, it is known that product value
emotional trust on the purchase of products by the audience watching
and product style can be characterized by product price and product
the food category live broadcast, respectively. λ11 and λ21 are the
category, while perceived quality depends on consumers’ subjective
regression coefficients of the control variables, respectively; ε1 and ε2
feelings and cannot be expressed by objective data. Therefore, this study
are the residual terms.
adds product price (PP) as a control variable in the model for research,
and the meanings of the variables are shown in Table 8.
(3) Empirical Results Analysis of Econometric Model
Among the constructed variables, the dependent, independent and
control variables all involve two attributes, i.e., time and website cate­
This study used Stata/MP 13.0 software to analyze the data. The
gory (the online platform trading website where the online store of the
descriptive statistics of the number of bullet screen word crawled by
Online celebrity product is located). All variables are standardized so
each Online celebrity live broadcast website are shown in Table 9.
that the dependent, independent and control variables are left with a
Further descriptive statistics were performed for SN, PN and PP and
time dimension, and a more robust result can be obtained by using the
the results are shown in Table 10.
least squares method for regression.
The standardized data were put into equations (1) and (2) for
In summary, 2 models are established in this study.
regression analysis, and the model regression results are shown in
Model 1: Online celebrity recommended product search model
Table 11.
∑
2 ∑
2 ∑
2
SN = C0 + C1 + C2 + σ 11j JO • Ij + σ12j AR • Ij + σ 13j AD • Ij
4.6. Results discussion
j=0 j=0 j=0

∑
2
+ σ14j TR • Ij + λ11 PP + ε1 (3)
j=0 The results of Study 2 show that the emotions of other audiences who
are watching the same livestream can also influence consumers’ will­
Model 2: Online celebrity recommended product purchase model ingness to purchase products recommended by the livestream. Specif­
ically, for the effect of watching the cosmetics category livestream on
product search, the coefficients of pleasure emotion, arousal emotion,

Table 9
Descriptive statistics of number of words for each dimension of information
Table 8 source characteristics for three product types (days).
Meaning of measured variables.
Online Emotional Number Median SD Min. Max.
Variable Type Variables Meaning of Variables Celebrity Characteristics of words
Type Mean
Dependent SN Search number for products during the data
variable collection period Cosmetics JO 90.60 88.50 53.01 5 198
PN Purchas number of products during the data AR 550.98 559.50 316.25 36 1133
collection period AD 489.40 491.00 280.922 32 1008
Independent IJ Online Celebrity type, for cosmetics category, j TR 640.72 647.50 368.55 42 1311
variable takes the value of 0, for clothes category, j takes the Clothes JO 253.50 260.50 132.923 18 494
value of 1, for food category, j takes the value of 2 AR 349.43 359.50 184.44 24 676
JO Number of pleasure emotion words AD 305.82 313.50 162.14 23 600
AR Number of Arousal emotion words TR 412.51 427.00 218.10 29 811
AD Number of Admiration words Food JO 385.49 386.50 201.79 43 786
TR Number of Emotional trust words AR 252.52 249 132.90 30 498
Controlled PP Price of products searched and purchased by AD 293.20 295.50 128.22 76 497
Variables consumers TR 220.83 217 116.36 25 451

9
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Table 10
Descriptive statistics results.
Variables No. of Days Mean Median SD Min. Max.

SN（Cosmetics） 288 1983.57 2005.00 1138.10 133 3967

PN（Cosmetics） 288 247.57 250 113.50 63 446
PP（Cosmetics） 288 372.70 388 164.41 85 652
SN（Clothes） 288 1096.39 1163.50 575.74 82 2099
PN（Clothes） 288 229.20 236 57.58 128 329
PP（Clothes） 288 458.41 453 166.05 166 746
SN（Food） 288 771.08 765.50 403.06 91 1494
PN（Food） 288 96.32 95.50 40.13 29 169
PP（Food） 288 221.94 215.50 101.18 50 399

admiration emotion, were found to be 0.525, 1.183, and 1.209 respec­

Table 11
tively, and were statistically significant, while emotional trust was not
Regression results.
significant; for the effect of watching the food livestream on product
Variables SN PN purchase, the coefficients of pleasure emotion, arousal emotion, and
I0 − 1.397（5.787） 48.469***（1.274） admiration emotion were found to be 0.086, 0.109, and 0.073 respec­
I1 3.409（4.431） 118.845***（0.510） tively, and were statistically significant, while emotional trust was not
I2 − 9.434（5.096） 19.539***（1.408）
significant. Hence, H1b, H2b, and H3b were verified, while H4b was not.
JO⋅I0 0.879***（0.078） 0.079***（0.018）
JO⋅I1 1.352***（0.126） 0.130***（0.013） Therefore, the results show that for other audience members who
JOn⋅I2 0.525***（0.059） 0.073**（0.027） watched the food Online celebrity livestream, the collective effect of
AR⋅I0 1.531***（0.091） 0.149***（0.020） pleasure emotion, arousal emotion, and admiration emotion influenced
AR⋅I1 0.811***（0.089） 0.083***（0.009） others and increased the search and purchase behavior, while emotional
AR⋅I2 1.183***（0.089） 0.086***（0.018）
AD⋅I0 1.003***（0.108） 0.107***（0.024）
trust had no effect on consumer search and purchase intentionThis
AD⋅I1 0.899***（0.091） 0.093***（0.009） contradicts the findings of previous studies on advertising virtual
AD⋅I2 0.024**（0.010） 0.109***（0.028） spokespersons, which have found that trustworthiness is a key factor
TRt⋅I0 0.932***（0.262） 0.136*（0.059） influencing consumers’ feelings and decisions (Keeling et al., 2010). In
TRt⋅I1 0.465***（0.065） 0.046***（0.006）
this study, we argue that consumers are more likely to rationalize when
TR⋅I2 1.209***（0.094） 0.004（0.003）
PP 0.002*（0.001） 0.005（0.003） faced with food-related decisions (Moore and Konrath, 2015). There is a
R2 0.998 0.996 greater focus on the description and acquisition of objective information
(Rodrigo et al., 2018). Also, in psychological studies, it is shown that
when performing food-related information processing, consumers
admiration emotion, and emotional trust were found to be 0.879, 1.352, enable the cognitive decision-making system for rational analysis
1.003, and 0.932 respectively and were statistically significant; mean­ (Moody et al., 2010). Therefore, for enhancing the willingness to pur­
while, for the effect of watching the cosmetics category livestream on chase food products, consumers need deeper cognitive trust rather than
product purchase, the coefficients of pleasure emotion, arousal emotion, the shallow emotional trust triggered by the Online celebrities’ perfor­
admiration emotion, and emotional trust were 0.079, 0.149, 0.107, and mances alone.
0.136 respectively, and were statistically significant. Therefore, H1b,
H2b, H3b, and H4b were verified.
5. Conclusion implication and limitation
For the effect of watching the clothes livestream on product search,
the coefficients of pleasure emotion, arousal emotion, admiration
5.1. Research conclusion
emotion, and emotional trust were found to be 1.352, 0.811, 0.899, and
0.465 respectively, and were statistically significant; meanwhile, for the
This study investigated the influence of online celebrities in live­
effect of watching the clothes livestream on product purchase, the co­
streaming e-commerce on consumers’ purchase intention and its un­
efficients of pleasure emotion, arousal emotion, admiration emotion,
derlying mechanism based on the theory of emotional contagion.
and emotional trust were found to be 0.130, 0.083, 0.093, and 0.046
Through a survey for obtaining data and related analyses, we examined
respectively, and were statistically significant. Hence, H1b, H2b, H3b,
and verified that consumers are influenced with pleasure emotion,
and H4b were verified. In this circumstance, it was found that the au­
admiration emotion, arousal emotion, and emotional trust by online
diences watching the cosmetics and clothes categories of Online celeb­
celebrities during livestreams, which enhances their purchase in­
rity livestreaming through screen bullets, with the effect of pleasure
tentions. We developed a PYTHON crawler for obtaining screen bullets
emotion, arousal emotion, admiration emotion, and emotional trust can
during livestreams and applied an econometric model approach to
influence others’ search and purchase behaviors. Among them,
analyze the data sets. The results show that audiences are emotionally
emotional trust and arousal emotion are the most significant, followed
influenced by other audience members along with online celebrities,
by admiration emotion and pleasure emotion.
which has an effect on their purchasing behavior.
A possible reason is that online celebrities often demonstrate
insightful knowledge, through their own experience and professional
analysis, to show the function of the products and explain the eye- 5.2. Theoretical contributions
catching points with emotion, which will greatly reduce the perceived
risk of consumers for cosmetics and clothing products, and they set up In terms of theoretical advancement, first, previous research on
activities, such as a limited time flash sale, to interact with the audience. emotional contagion has mostly focused on social psychology (HKelly
Consumers will be influenced by the emotional trust, arousal emotion, and Barsade, 2001), and the only few applications in marketing have
admiration emotion, and pleasure emotion that are aroused, which in been on the influence of service personnel’s emotional changes on
turn, will enhance their purchasing behavior. customers’ emotions and their purchase behavior in offline service
For the effect of watching the food livestream on online product contexts (Du et al., 2011). This study applies the emotion contagion
search, the coefficients of pleasure emotion, arousal emotion, and theory to understand the emotion-related issues in livestreams, and it
shows that consumers are influenced with pleasure emotion, admiration

10
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

emotion, arousal emotion, and emotional trust during the livestreams, livestreams as a tool for product promotion should be taken seriously,
which enhances their purchase intentions. Thus, it has extended the but at the same time, the importance of improving product quality
application domain of emotion contagion theory. should be not ignored.
In addition, livestreams of different product types have been
considered. It was found that for the food and clothing categories, 5.4. Research limitations and future directions
pleasure emotion, arousal emotion, admiration emotion, and emotional
trust can influence purchase behavior. For the food category, only This study has several limitations. First, we explored only the influ­
pleasure emotion, arousal emotion, and admiration emotion have such ence of positive emotions on consumers’ purchase behavior during the
an effect. For other audience members watching the food livestream, the livestreams, but we did not explore negative emotions such as fear,
pleasure emotion, arousal emotion, and admiration emotion can fright, sadness, and so forth. Previous studies on emotions have found
emotionally influence consumers’ online searches and purchase that negative emotions may influence consumers to participate in
behavior, but emotional trust cannot; possibly due to the special nature compensatory consumption, and subsequent studies can further explore
of food, consumers rely more on rational thinking to make decisions the influence of negative emotions during livestreams. Secondly, this
(Moody et al., 2010). study focused only on the emotional perspective of livestreams but did
Compared to traditional offline shopping (Beck and Dominique, not consider the influence of the cognitive perspective on consumer
2016) and celebrity endorsement advertising (Mccormick, 2016), the behavior, which can be further explored by organically integrating the
most significant feature of livestreaming e-commerce is the real-time two together in the future. Moreover, a contextual questionnaire was
multi-directional interaction through screen bullets, which directly used in Study 1, and the subjects’ purchase intention may not actually
connects audiences with online celebrities as well as with other audi­ translate into purchase behavior in reality. Future research can consider
ences, constituting a main content of the livestreams. As a result, con­ using consumer laboratory experiments and field experiments to further
sumers are not only influenced by the online celebrities during the verify the intrinsic mechanism of livestreaming and to better understand
process of watching livestreams but also by other audiences. This study, consumers’ purchasing behavior. Finally, in Study 2, we selected only
developed from the perspective of multiparty emotional contagion, can three product types, namely, the cosmetics, clothing, and food cate­
consider online celebrities themselves and other audiences as a joint gories, so future research could examine other product types and may
whole, arguing that the performances of online celebrities will yield other, more interesting conclusions.
emotionally influence consumers, and audiences influenced by online
celebrities can also indirectly emotionally influence other audiences in Availability of data and material
terms of purchase intentions.
If the editor needs any data in our research, please contact with jj
5.3. Managerial implications [email protected].

In terms of the direct contagion effect, and the results regarding Authors’ contributions
online celebrities’ role and performance, they may consider paying more
attention to their own emotion-related elements in their performances Lu Meng is responsible for the construction and later revision of this
since they have a direct influence on audience purchase behavior. Also, paper. Shen Duan is responsible for theoretical literature review, article
consumers’ emotional needs should be considered comprehensively, writing and adjusting. Yijun Zhao is responsible for article language
and corresponding combined programs could be formulated. touch-up and data collection. Kevin Lu is responsible for sorting out the
In terms of indirect contagion, online celebrities should increase data and related auxiliary work.
their interaction with audiences and inspire the generation of more
screen bullets. Specifically, this can be done through a more in-depth Fund
and down-to-earth way of marketing and event organization. To be
specific, online celebrities can interact with the audience more This work was supported by the National Natural Science Foundation
frequently by reading audiences’ bullet screen comments. They can also of China NSFC (No.72002153).
make segmentation for the audience watching the live broadcast. Under
the premise of professional explanation of product-related knowledge, Acknowledgements
for male audiences, more interactive lottery draw sessions can be set up,
while for female audiences, more funny and pleasant atmosphere can be The authors thank the JRCS review team for their excellent guidance
created, so that consumers in the virtual space can produce strong throughout the review process.
sensory impact and emotional experience.
For consumers, when viewing livestreams, it is necessary for them to Appendix A
keep a clear mind, to fully consider their actual needs, to promote
rational consumption rather than impulsive consumption, and to choose “Bullet screen”, or “dan’mu” in Chinese, is an emerging new feature
credibility and cost-effective products in order to have a pleasant on online video sites in China and Japan, which allows real-time com­
shopping experience. Also, it can be a learning opportunity, for asking ments from viewers to fly across the screen like bullets.
questions and for sharing and exchanging information, opinion, and Bilibili, a leading Chinese video streaming website famous for its
comments with other audience members. danmu ("bullet screens" in Chinese) service that lets viewers post or send
Thirdly, for products and livestreaming e-commerce companies, each other text messages while watching a film or show, became one of
online celebrities can indeed bring a high number of online product the most popular text messaging platforms in China in 2017, with over 1
searches and a high turnover rate. However, the high traffic brought by billion commentary subtitles.
online celebrities may not equate to brand value let alone sustainable
branding. The enhancement of brand value lies in its own brand con­ References
struction, such as assets, information technology, channel construction,
and word-of-mouth management, as well as experiential marketing, and Barsade, S.G., Coutifaris, C.G., Pillemer, J., 2018. Emotional contagion in organizational
so forth. Livestreaming is only a promotional tool to arouse consumers’ life. Res. Organ. Behav. 38, 137–151.
Becker, B.E., Luthar, S.S., 2007. Peer-perceived admiration and social preference:
emotional impulses. Only on the basis of good brand building can this contextual correlates of positive peer regard among suburban and urban adolescents.
approach to promotion be meaningful for brands. Therefore, the role of J. Res. Adolesc. 17 (1), 117–144.

11
L.(M. Meng et al. Journal of Retailing and Consumer Services 63 (2021) 102733

Beck, M., Dominique, C., 2016. I virtually try it i want it! virtual fitting room: a tool to Mccormick, K., 2016. Celebrity endorsements: influence of a product-endorser match on
increase on-line and off-line exploratory behaviour, patronage and purchase millennials attitudes and purchase intentions. J. Retailing Consum. Serv. 32 (9),
intentions. J. Retailing Consum. Serv. 40 (1), 279–286. 39–45.
Chen, Y.H., Chen, M.C., Keng, C.J., 2020. Measuring online live streaming of perceived Mian, S.A., Hattab, H.W., 2013. How individual competencies shape the entrepreneur’s
servicescape: scale development and validation on behavior outcome. Internet Res. social network structure: evidence from the MENA region. Int. J. Bus. Glob. 11 (4),
30 (3), 737–762. 399–412.
Cheng, F.F., Wu, C.S., Yen, D.C., 2009. The effect of online store atmosphere on Monsuwé, T.P., Gellaert, B.G.C., Ruyter, K., 2004. What drives consumers to shop online?
consumer’s emotional responses - an study of music and colour. Behav. Inf. Technol. A literature review. Int. J. Serv. Ind. Manag. 15 (1), 102–121.
28 (4), 323–334. Moody, W., Kinderman, P., Sinha, P., 2010. An exploratory study: relationships between
Chua, R.Y.J., Ingram, P., Morris, M.W., 2008. From the head and the heart: locating trying on clothing, mood, emotion, personality and clothing preference. J. Fash.
cognition- and affect-based trust in managers’ professional networks. Acad. Manag. Mark. Manag. 14 (1), 161–179.
J. 51 (3), 436–452. Moore, D.J., Konrath, S., 2015. I can almost taste it:" why people with strong positive
Chuah, H.W., Yu, J., 2021. The future of service: the power of emotion in human-robot emotions experience higher levels of food craving, salivation and eating intentions.
interaction. J. Retailing Consum. Serv. 61 (3), 102551. J. Consum. Psychol. 25 (1), 42–59.
DelVicario, M., Bessi, A., Zollo, F., Petroni, F., Scala, A., Caldarelli, G., Stanley, E., Oliver, R.L., Rust, R.T., Varki, S., 1997. Consumer delight: foundations, findings, and
Quattrociocchi, W., 2016. The spreading of misinformation online. Proc. Natl. Acad. managerial insight. J. Retailing 73 (3), 311–336.
Sci. USA 113 (3), 554–559. Park, H.J., Lin, L.M., 2020. The effects of match-ups on the consumer attitudes toward
Dodds, W.B., Monroe, K.B., Grewal, D., 1991. Effect of price, brand and store information online celebrities and their live streaming contents in the context of product
on buyers’ products evaluations. J. Mark. Res. 28 (3), 307–319. endorsement. J. Retailing Consum. Serv. 52, 1–6.
Donovan, R.J., Rossiter, J.R., Marcoolyn, G., 1994. Store atmosphere and purchasing Peng, L., Cui, G., Chung, Y., Zheng, W., 2020. The faces of success: beauty and ugliness
behavior. J. Retailing 70 (3), 283–294. premiums in e-commerce platforms. J. Mark. 84 (4), 67–85. https://doi.org/10.11
Du, J., Fan, X., Feng, T., 2011. Multiple emotional contagions in service encounters. 77/0022242920914861.
J. Acad. Market. Sci. 39 (3), 449–466. Pihlström, M., Brush, G.J., 2008. Comparing the perceived value of information and
Fan, R., Xu, K., Zhao, J., 2017. An agent-based model for emotion contagion and entertainment mobile services. Psychol. Market. 25 (8), 732–755.
competition in online social media. Physica A 495, 245–259. Rodrigo, P., Khan, H., Ekinci, Y., 2018. The determinants of foreign product preference
Fei, M., Tan, H., Peng, X., Wang, Q., Wang, L., 2020. Promoting or attenuating? an eye- amongst elite consumers in an emerging market. J. Retailing Consum. Serv. 46 (1),
tracking study on the role of social cues in e-commerce livestreaming. Decis. Support 139–148.
Syst. 142 (1–2), 113466. Russell, J.A., 1980. A circumplex model of affect. J. Pers. Soc. Psychol. 39 (6),
Habibi, M.R., Laroche, M., Richard, M.O., 2014. The roles of brand community and 1161–1178.
community engagement in building brand trust on social media. Comput. Hum. Russell, J.A., Anna, W.A., Mendelsohn, G.A., 1989. Affect grid: a single-item scale of
Behav. 37 (37), 152–161. pleasure and arousal. J. Pers. Soc. Psychol. 57 (3), 493–502.
Hatfield, E., Cacioppo, J., Rapson, R.L., 1994. Emotional Contagion. Cambridge Sun, Y., Shao, X., Li, X., Guo, Y., Nie, K., 2019. How live streaming influences purchase
University Press, New York. https://doi.org/10.1017/CBO9781139174138. intentions in social commerce: an it affordance perspective. Electron. Commer. Res.
Hilvert-Bruce, Z., Neill, J.T., Sjöblom, M., Hamari, J., 2018. Social motivations of live- Appl. 37, 1–12.
streaming viewer engagement on Twitch. Comput. Hum. Behav. 84, 58–67. Teh C, 2021 April 16. “Meet China’s ’Lipstick King,’ an Outspoken 28-Year-Old
HKelly, J.R., Barsade, S.G., 2001. Mood and emotions in small groups and work teams. Ecommerce Streamer Who Fans Adore and Brands Fear”. Business Insider Australia.
Organ. Behav. Hum. Dec. 86 (1), 99–130. https://www.businessinsider.com.au/austin-li-jiaqi-chinas-lipstick-king-onlin
Hou, F., Guan, Z., Li, B., Chong, A.Y.L., 2019. Factors influencing people’s continuous eshopping-taobao-2021-3.
watching intention and consumption intention in live streaming: evidence from Trivedi, J.P., Sama, R., 2020. The effect of influencer marketing on consumers’ brand
China. Internet Res. 30 (1), 141–163. admiration and online purchase intentions: an emerging market perspective.
Immordino-Yang, M.H., McColl, A., Damasioa, H., Damasio, A., 2009. Neural correlates J. Internet Commer. 19 (1), 103–124.
of admiration and compassion. Neuroscience 106 (19), 8021–8026. Wohlfeil, M., Whelan, S., 2012. “Saved!” by Jena Malone: an introspective study of a
Keeling, K., McGoldrick, P., Beatty, S., 2010. Avatars as salespeople: communication consumer’s fan relationship with a film actress. J. Bus. Res. 65 (4), 511–519.
style, trust and intentions. J. Bus. Res. 63 (8), 793–800. https://doi.org/10.1016/j.jbusres.2011.02.030.
Kim, H.S., Kim, M., 2020. Viewing sports online together? psychological consequences You, S., Robert Jr., L.P., 2018. Emotional attachment, performance, and viability in
on social live streaming service usage. Sport Manag. Rev. 6 (2), 1–12. teams collaborating with embodied physical action (EPA) robots. J. Assoc. Inf. Syst.
Kramer, A.D., Guillory, J.E., Hancock, J.T., 2014. Experimental evidence of massive－ Online 19 (5), 42.
scale emotional contagion through social networks. Proc. Natl. Acad. Sci. Unit. States Yu, S., Hu, Y., 2020. When luxury brands meet China: the effect of localized celebrity
Am. 111 (24), 8788–8790. endorsements in social media marketing. J. Retailing Consum. Serv. 54, 102010.
Ladhari, R., 2007. The effects of consumption emotions on satisfaction and word-of- Zhang, K.Z., Cheung, C.M., Lee, M.K., 2014. Examining the moderating effect of
mouth communications. Psychol. Market. 24 (12), 1085–1108. inconsistent reviews and its gender differences on consumers’ online shopping
Magnus, S., Eeva-Liisa, O., Teck, M.T., 2021. The happy virtual agent and its impact on decision. Int. J. Inf. Manag. 34 (2), 89–98.
the human customer in the service encounter. J. Retailing Consum. Serv. 59 (3), Zhou, J., Zhou, J., Ding, Y., Wang, H., 2019. The magic of danmaku: a social interaction
102401. perspective of gift sending on live streaming platforms. J. Electron. Commer. Res. A.
34, 1–9.

12