Engaging in Dialogues: The Impact of Comment Valence and Influencer-Viewer Interaction on the Effectiveness of YouTube Influencer Marketing

Abstract

YouTube is a popular social marketing platform. Marketers or advertisers can collaborate with a YouTube influencer to present marketing messages. However, negative user-generated comments may affect the effectiveness of message delivery. Thus, one pretest and two main studies were conducted to investigate the influence of negative comments on consumers and the strategy to combat the negativity. The first study examined the influence of comment valence on product attitude and perceived trustworthiness of influencer. The second study examined how the increased frequency of influencer–viewer interaction mitigated the damage inflicted by negative comments. The findings of the studies reveal that negative comments have a strong influence on consumers. However, if an influencer is actively replying to negative comments, the negative influence is likely to be mitigated. Theoretical and practical contributions of the studies were discussed.

Full text

Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=ujia20
Journal of Interactive Advertising
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ujia20
Engaging in Dialogues: The Impact of Comment
Valence and Influencer-Viewer Interaction on the
Effectiveness of YouTube Influencer Marketing
Min Xiao
To cite this article: Min Xiao (2023): Engaging in Dialogues: The Impact of Comment Valence
and Influencer-Viewer Interaction on the Effectiveness of YouTube Influencer Marketing, Journal of
Interactive Advertising, DOI: 10.1080/15252019.2023.2167501
To link to this article: https://doi.org/10.1080/15252019.2023.2167501
Published online: 25 Jan 2023.
Submit your article to this journal
View related articles
View Crossmark data

Engaging in Dialogues: The Impact of Comment Valence and Influencer-
Viewer Interaction on the Effectiveness of YouTube Influencer Marketing
Min Xiao
Elliott School of Communication, Wichita State University, Wichita, Kansas, USA
ABSTRACT
YouTube is a popular social marketing platform. Marketers or advertisers can collaborate
with a YouTube influencer to present marketing messages. However, negative user-gener-
ated comments may affect the effectiveness of message delivery. Thus, one pretest and two
main studies were conducted to investigate the influence of negative comments on con-
sumers and the strategy to combat the negativity. The first study examined the influence of
comment valence on product attitude and perceived trustworthiness of influencer. The
second study examined how the increased frequency of influencer–viewer interaction miti-
gated the damage inflicted by negative comments. The findings of the studies reveal that
negative comments have a strong influence on consumers. However, if an influencer is
actively replying to negative comments, the negative influence is likely to be mitigated.
Theoretical and practical contributions of the studies were discussed.
KEYWORDS
Influencer marketing;
negativity bias; the MAIN
model; user-generated
comments; YouTuber
Product-related YouTube videos are resources that
help consumers make informed purchase decisions
(Ellet 2018; Swant 2015). Many product-related
YouTube videos are created by so-called influencers
(Lorenz 2019). In those videos (e.g., product review,
haul, unboxing), products are often the focus of the
video content and viewers can comment on the con-
tent. However, not all comments are positive.
Negative comments, posted by unsatisfied consumers
or by Internet trolls, impair the reputation of an influ-
encer and the image of the product presented in the
influencer’s video.
Empirical studies of influencer marketing have
investigated topics such as parasocial interaction (Lee
and Watkins 2016; Jin and Ryu 2020; Yuan and Lou
2020) and sponsorship disclosure (Boerman 2020;
Boerman and Reijmersdal 2019). However, the influ-
ence of negative user-generated comments, which may
impair consumer perception of a product and an
influencer, has been largely overlooked. Thus, a goal
of the current research is to examine the psychological
influence of negative comments on consumers in the
realm of YouTube influencer marketing from the the-
oretical perspective of the negativity bias.
Another goal of the study is to ascertain the effect-
iveness of leveraging influencer–viewer interaction
frequency as a strategy to reduce the impact of nega-
tive comments. Some YouTube influencers have
undertaken strategies, such as disabling the comment
section, deleting negative comments, or ignoring nega-
tive comments, as ways to contain the negativity.
However, these strategies may not be effective and
some of them may even have an adverse effect. In
contrast, if an influencer confronts negative comments
by replying to these comments at a frequent rate, the
influencer may provide additional information that
can help resolve viewers’concerns. More importantly,
frequent influencer–viewer interactions are conducive
to create a positive image of an influencer as a
friendly and trustworthy person. Such a positive per-
ception of an influencer in the eyes of consumers can
be regarded as an outcome of heuristic information
processing.
Empirical studies have examined psychological heu-
ristics from various theoretical perspectives (Chaiken
1980; Petty and Cacioppo 1984; Sundar 2008).
According to these studies, information cues, such as
the valence of social media posts and the number of
posts (Waddell and Sundar 2017), activate the usage
of heuristics. A valuable theoretical framework to help
scholars investigate the influence of heuristic and
information cues on digital media consumers is the
CONTACT Min Xiao min.xiao@wichita.edu Elliott School of Communication, Wichita State University, 1845 Fairmount Street, Wichita, KS
67260, USA.
Min Xiao (PhD, University of Florida) is an assistant professor of communication at Wichita State University.
ß2023 American Academy of Advertising
JOURNAL OF INTERACTIVE ADVERTISING
https://doi.org/10.1080/15252019.2023.2167501

MAIN model. Specifically, empirical evidence obtained
from the literature on heuristic processing was utilized
in conjunction with the negativity bias to explain the
effects unearthed by the current research. The proposed
explanatory mechanism is not only a bridge that closes
the conceptual gap between user-generated comments
and consumer perceptions but also a theoretical foun-
dation that explains how the increased frequency of
influencer–viewer interaction mitigates the influence of
negative comments on consumers.
In general, the goals of this research study are to
examine how negative user-generated comments affect
influencer marketing effectiveness on YouTube and
how the increased frequency of influencer–viewer inter-
action helps mitigate the negative influence. One pre-
test and two main studies (i.e., experiments) were
conducted to investigate the issue. The pretest and first
experiment examined how comment valence would
affect product attitude and perceived trustworthiness of
influencer. The second experiment investigated whether
the increase in the frequency of influencer–viewer
interaction would mitigate the negative influence of
user-generated comments on consumer perception.
Anecdotal evidence has documented the influence of
negative comments and influencer–viewer interaction
on consumers, yet no empirical research has been con-
ducted to systematically investigate the validity of such
observations. The current research is the first attempt
to validate personal observations empirically and sys-
tematically in the context of influencer marketing. The
findings of the current research would enrich the body
of knowledge about online human interactions and
heuristic information processing. The findings would
also offer managerial implications to brands and mar-
keters that help them better conduct influencer market-
ing not only on YouTube but also on other social
media platforms.
Literature Review
Influencer Marketing
The number of research studies that examine influen-
cer marketing is growing. Extant studies have investi-
gated the topic from an array of angles. Lou and
Yuan (2019) proposed the social media influencer
model and discovered that factors, such as inform-
ativeness of content, influencer trustworthiness, and
similarity between influencers and followers, affected
consumers’trust in an influencer’s branded posts.
Childers, Lemon, and Hoy (2019) interviewed 19
advertising professionals in the United States and
asked about their opinion of influencer marketing.
Most of the interviewed professionals considered
influencer marketing an uncharted territory.
According to empirical research, consumers perceived
an Instagram influencer in a more favorable way
when the influencer had a large number of followers
than when the influencer had a small number of fol-
lowers (De Veirman, Cauberghe, and Hudders 2017).
Jang et al. (2021) examined travel influencers and dis-
covered that the level of consumer engagement with
the influencer significantly affects the effectiveness of
advertising message delivery. Lee and Kim (2020)
uncovered a strong influence of brand credibility on
message credibility, attitude toward advertisement,
purchase intention, and electronic word-of-mouth
(eWOM) intention in the context of Instagram influ-
encer marketing. Schouten, Janssen, and Verspaget
(2020) investigated the impact of influencers and
celebrities on advertising effectiveness. Their findings
suggest that consumers trust influencers more than
celebrities, and perceived trustworthiness of influencer
mediates the relationship between the type of endors-
ers (i.e., influencer or celebrity) and advertising effect-
iveness (i.e., attitude and purchase intention). A few
academic research projects specifically focused on
studying influencer marketing on YouTube. For
instance, Boerman and Reijmersdal (2019) discovered
that sponsorship disclosure presented in an influen-
cer-made YouTube video enhanced brand memory
but decreased the intention to own the product. The
reviewed studies thus far have advanced scholars’
understanding of influencer marketing, but the foci of
the studies were not on user-generated comments or
psychological heuristics.
The Negativity Bias in eWOM
The negativity bias is a notion that negative events are
weighed as more potent and perceived as more salient
than positive events (Ito et al. 1998; Rozin and
Royzman 2001). Some scholars attribute the cause of
the bias to the loss-aversion mentality (Tversky and
Kahneman 1991), while others consider the negativity
bias a result of positive-negative asymmetry because
the occurrence of negative events is much rarer than
positive events (Lewicka, Czapinski, and Peeters
1992). Scholars who plan to investigate the impact of
negative user-generated comments from a perspective
of the negativity bias in the context of social media
influencer marketing may obtain important informa-
tion from reviewing empirical studies about eWOM
and product reviews in a media marketing context.
2 M. XIAO

Waddell and Sundar (2017) examined how the
valence of tweets affected viewer perception of a tele-
vision program, and their findings were consistent
with the empirical evidence obtained from studies of
negativity bias: Negative tweets were more impactful
than positive tweets in cueing the bandwagon heuris-
tic. Hong and Pittman (2020) researched how the
review score, review count, and review valence
affected the perceived credibility of product reviews,
and they found that positive reviews enhanced study
participants’trust in the star ranking system and that
negative reviews reduced participants’trust in the
same system and encouraged them to focus more on
the number of reviews. Other than studies conducted
by communication scholars, empirical research con-
ducted by marketing scholars has documented the
psychological impact of negative reviews on consum-
ers. However, the findings of the studies are inconsist-
ent. Vermeulen and Seegers (2009) found that positive
hotel reviews had a significant impact on consumer
behavior, whereas negative reviews had little or no
impact. In contrast, Sparks and Browning (2011) dis-
covered that consumers were more likely to be influ-
enced by negative reviews than by positive ones,
especially when the general tone of reviews was nega-
tive. Filieri, Raguseo, and Vitari (2021) analyzed more
than 9,000 Tripadvisor reviews and discovered an
overwhelming influence of negative hotel ratings on
consumers’perception of high-quality hotels.
However, Wu (2013) analyzed reviews on
Amazon.com and found no evidence of the negativity
bias when review quality was controlled. Chen and
Lurie (2013) examined Yelp.com reviews and uncov-
ered mixed results: Negative reviews were more valu-
able to consumers than positive reviews in general,
but the negativity bias was absent if temporal contigu-
ity cues were controlled.
Overall, the research studies reviewed thus far sug-
gest a significant association between eWOM valence
and consumer perception. However, the inconsistent
findings unearthed by some of the studies obfuscated
the interpretation of how negative comments on
YouTube may affect viewers’attitude toward a prod-
uct presented in a video. Although inconsistency in
empirical findings exists, the inconsistency is likely
caused by the variance in research methods and
research contexts (e.g., differences between groups of
consumers). Thus, an empirical investigation is
needed to verify the validity of the existing empirical
findings. Hence, the theoretical foundation of this
study is rooted in the canonical literature of the nega-
tivity bias and the following hypothesis was posited
based on the core principles of the negativity bias
concept proposed in the key literature (Ito et al. 1998;
Rozin and Royzman 2001; Tversky and Kahneman
1991).
H
1
Negative comments accompanying a YouTube
video will lead viewers to have a significantly more
negative attitude toward the product presented in the
video than the product presented in a video with posi-
tive comments or no comment.
The MAIN Model
The MAIN model addresses the importance of how
an affordance (i.e., action possibilities) conveys infor-
mation cues that trigger an individual to use a heuris-
tic (Evans et al. 2017; Sundar 2008). The name of the
model, MAIN, is an acronym, representing the affor-
dances of Modality (M), Agency (A), Interactivity (I),
and Navigability (N). Media consumers need to spend
extra time and energy in assessing the quality of infor-
mation because the overabundance of digital media
content obscures the sources of information and the
means of information dissemination (Fogg, Cuellar,
and Danielson 2002). However, individuals are also
known as “cognitive misers,”and they tend to expend
as little cognitive effort as possible to process informa-
tion (Fiske and Taylor 1991).
The academic definition of heuristic varies, but the
essence of the construct is the same. Gigerenzer and
Gaissmaier (2011) defined heuristic as “a strategy that
ignores part of the information, with the goal of mak-
ing decisions more quickly, frugally, and/or accurately
than more complex methods”(454). Sundar (2008)
defined heuristic as a mental generalization of know-
ledge or experience that expedites information proc-
essing. All-in-all, a psychological heuristic, which is
often triggered by information cues, is a mental short-
cut that helps individuals evaluate information in a
time- and energy-conservative manner. As explained
earlier in the manuscript, heuristic and information
cue are two different constructs; heuristic denotes a
cognitive shortcut, while information cue represents
the trigger of the heuristic. In this study, comment
valence and the frequency of influencer–viewer inter-
action are regarded as information cues that activate
heuristic processing. Previous studies, which employed
the MAIN model as the theoretical framework, have
predominantly focused on examining the bandwagon
heuristic (e.g., Lee, Atkinson, and Sung 2020; Sundar,
Oeldorf-Hirsch, and Xu 2008;Xu2013).
JOURNAL OF INTERACTIVE ADVERTISING 3

Perceived Trustworthiness of Influencer
The construct of information source trustworthiness
denotes the apparent integrity of the source or the
perceivers’confidence in the source to communicate
valid and honest assertions (Hovland, Janis, and
Kelley1953, 21; McGinnies and Ward 1980). In other
words, the construct represents the reliability or
trustworthiness of an information source in the eyes
of the perceivers, and this apparent trustworthiness is
not necessarily related to the actual trustworthiness
of the source (Gass and Seiter 2011). Note that the
study is focusing on examining the perceived trust-
worthiness of influencer, and the actual trustworthi-
ness of an influencer is not a focus of the current
research.
The perceived trustworthiness of an information
source has been studied in various contexts as an
antecedent or independent variable that influences
certain attitudinal or behavioral outcomes. In the con-
text of marketing and advertising on traditional media
platforms, source trustworthiness has been examined
as a factor in predicting brand attitude and brand
beliefs (Yoon, Kim, and Kim 1998). In the context of
influencer marketing, perceived trustworthiness of an
influencer has been reported as an antecedent corre-
lated with brand image (Mettenheim and Wiedmann
2021) and brand trust (Lou and Yuan 2019).
However, what precedes, or triggers, perceived trust-
worthiness of an information source has not garnered
enough attention from academic scholars.
In the current study, the perceived trustworthiness
of an influencer is regarded as a perceptional outcome
of heuristic processing. In many scenarios, consumers
use circumstantial evidence or information cues, such
as the number of comments or overall comment
valence, to help them evaluate a person or an entity.
When watching videos produced by YouTube influ-
encers, consumers are likely to ignore part of the
information (e.g., the actual trustworthiness of a
source) and use subjective impressions to make quick
decisions (Gigerenzer and Gaissmaier 2011; Sundar
2008). One such decision is the perceived trustworthi-
ness of an influencer.
A specific psychological mechanism, the band-
wagon heuristic, can explain how user-generated com-
ments influence perceived trustworthiness of an
influencer. Many empirical studies have examined the
bandwagon heuristic from the news consumption
perspective. Knobloch-Westerwick et al. (2005) discov-
ered that individuals were more likely to read
news articles published by a source that was recom-
mended by a large number of readers. In terms of
user-generated comments accompanying an online
news article, comments were reported to influence
perceived news credibility (Waddell 2018) and to
shape perceived public opinion (Lee, Atkinson, and
Sung 2020) through the influence of the bandwagon
heuristic. Some studies examined the bandwagon
heuristic in the context of online shopping and digital
content consumption. Sundar, Oeldorf-Hirsch, and Xu
(2008) discovered that information cues, such as
aggregated product review scores and sales ranking,
influenced a consumer’s bandwagon perception and
affected purchase intention. Fu and Sim (2011)
unearthed a statistically significant association between
the existing viewership count of a YouTube video and
the future viewership count of the video. Waddell and
Sundar (2017) examined how the valence of four
tweets affected viewer perception of a television pro-
gram, and their findings were consistent with the
empirical evidence obtained from studies of the nega-
tivity bias: Negative tweets were more impactful than
positive tweets in activating the bandwagon heuristic.
Based on the review of literature about the band-
wagon heuristic, one can infer that the bandwagon
heuristic is a psychological mechanism that shapes
how user-generated comments influence consumer
perception of influencer trustworthiness.
On YouTube, a video-centric social platform, an
influencer is the source of information. Various infor-
mation cues may affect how viewers perceive the
video as well as the influencer. Positive comments are
usually consistent with what an influencer has pre-
sented in a video (e.g., product review or unboxing),
while negative comments may be inconsistent with
the information presented in a video. Due to the
negativity bias phenomenon, consumers may be more
attentive when they are reading negative comments
than when they are reading positive comments. Thus,
the extent of negative comments’influence on per-
ceived trustworthiness of influencer is likely to be
stronger than that of positive comments’influence on
the same construct.
In fact, existing research has reported findings that
corroborated the conjecture proposed in the current
research. Naab et al. (2020) studied online news read-
ers and discovered that critical comments (i.e., nega-
tive comments) significantly reduced perceived news
credibility. Such a negative impact would be mitigated
if other users responded with counterarguments.
K€
umpel and Unkel (2020) studied German news-
readers and unearthed a significant negativity bias
effect regarding the impact of comment valence on
perceived journalistic quality. At the intersection of
4 M. XIAO

the negativity bias and the bandwagon heuristic,
Waddell (2020) discovered that negative comments
exerted a strong influence on perceived news article
credibility via the mediation of bandwagon perception.
Hence, the following hypothesis was posited based on
empirical evidence.
H
2
Negative comments underneath a YouTube video
will lead viewers to perceive the influencer who pro-
duced the video as less trustworthy than when the
viewers are exposed to a video accompanied by posi-
tive comments or no comments.
Extant research has revealed perceived influencer
trustworthiness/credibility as a key mediator or mod-
erator of information processing outcomes. Saima and
Khan (2020) investigated influencer marketing in
India. They discovered that the credibility of an influ-
encer mediated the correlation between the independ-
ent variables, such as information quality and
entertainment value, and the dependent variable, pur-
chase intention. Xiao, Wang, and Chan-Olmsted
(2018) found that the perceived trustworthiness of
influencer, social influence, argument quality, and the
level of information involvement were correlated with
perceived information credibility and that perceived
information credibility positively predicted the attitude
toward a brand and an influencer. Reinikainen et al.
(2020) focused on examining user-generated com-
ments in the context of influencer marketing. Their
findings suggest that a parasocial relationship positively
influences the perceived credibility of an influencer,
and the correlation between the two variables is moder-
ated by comments; moreover, perceived influencer
credibility positively influences brand trust and pur-
chase intention. Other studies (e.g., Hayes and Carr
2015; Heinbach, Ziegele, and Quiring 2018)haveexam-
ined the credibility of information source (e.g., an
influencer) as a moderator that moderates the correl-
ation between message exposures and message process-
ing outcomes. However, the findings of the moderation
effects are inconsistent. Thus, based on empirical evi-
dence, one can infer that the construct, perceived trust-
worthiness of influencer, is likely to mediate the
relationship between information exposure and infor-
mation processing outcomes (e.g., attitude).
The interaction between influencers and viewers on
YouTube is manifested as the comment-and-reply
interaction in the comment section. The frequency of
such interactions is likely to be a factor influencing
consumers’opinions on products and influencers. The
efficacy of the interaction’s influence on consumers
can be discussed from four perspectives.
First, the frequency, or the number of times, an
influencer replies to user-generated comments is likely
to be perceived as an information cue: an element
that triggers the use of heuristics. Petty and Cacioppo
(1984) found that if an individual processed a message
heuristically or peripherally, the individual might sim-
ply count the number of arguments (i.e., information
cue) provided in a message and reasoned that the
greater the number of arguments, the better the qual-
ity of the arguments (i.e., heuristic). Empirical studies
also suggest that the use of information cues and heu-
ristics is not limited to heuristic or peripheral process-
ing (Chaiken and Maheswaran 1994; Petty and Brinol
2012). When an individual performs effortful informa-
tion processing, heuristics and cues are utilized by the
individual to further expedite processing efficiency
(Chaiken 1980).
Second, the comment section is a type of techno-
logical affordance, and the comment-and-reply inter-
action between viewers and influencers is likely to be
an information cue built on the affordance. The
increased frequency of influencer–viewer interactions
is likely to induce a heuristic called the contingency
heuristic. The contingency heuristic is defined as one’s
perception of the interdependency of messages in the
information exchange process (Sundar et al. 2016). A
clear perception of contingency helps an individual
perceive the information exchange process as unique,
timely, and reliable (Sundar 2008). The perceived con-
tingency of discourse is mainly influenced by the con-
tent or context of communication/interaction between
two or more parties. One should also note that the
increased frequency of interactions is likely to enrich
the content or context of the communication/interac-
tion. Hence, frequent interactions between YouTube
influencers and viewers may lead viewers of the video
to perceive their information consumption experience
as unique, timely, reliable, and informative.
Third, confirmation bias may play a role in affect-
ing consumers. Videos posted by an influencer on
YouTube are usually professionally made (Lorenz
2019). The aesthetic quality of these videos can be
assured, and the information presented in the video
may seem to be of high quality. Thus, viewers are
likely to have a positive first impression of the video
content. However, the positive impression may shift
as viewers of the video scroll down to read negative
user-generated comments posted by unsatisfied con-
sumers, Internet trolls, or even the influencer’s busi-
ness competitors. Such a shift is likely to cause an
individual to experience discomfort that drives the
individual to seek additional information that
JOURNAL OF INTERACTIVE ADVERTISING 5

confirms the established beliefs (Festinger 1962;
Nickerson 1998). The confirmation bias explanation is
plausible because participants in this research study
were directed to watch the influencer-produced video
before reading any other information listed in the
comment section.
Confirmation bias is the tendency of an individual
to seek and place greater weight on information that
confirms or supports an individual’spreexistingbeliefs
(Klayman 1995). In the digital era, media consumers
tend to evaluate online information that is consistent
with their established opinion more favorably than
information that is inconsistent with their established
opinion (Fischer et al. 2005). Metzger, Flanagin, and
Madders (2010) discovered that media consumers avoid
consuming information contradictory to their preexist-
ing beliefs. Knobloch-Westerwick, Mothes, and Polavin
(2020) unearthed the presence of confirmation bias in
news consumption that readers chose to consume news
articles that were consistent with their political lean-
ings. Metzger and Flanagin (2013) placed confirmation
bias in the context of psychological heuristics and
coined the term self-confirmation heuristic—a tendency
of one to perceive information as credible or not cred-
ible depending on whether the information confirms or
challenges one’s established beliefs.
Similar to what has been discussed in the section of
contingency heuristic, the content or context of the
interaction is the key to forming one’s opinion, while
the frequency of the interaction is an element that
enriches the content or context of the interaction.
When an influencer replies to negative comments, the
influencer addresses concerns raised in these com-
ments at least to some extent. The input provided by
an influencer helps the viewer realign the perceived
criticism of a video with the established positive
impression of the same video. The more frequently an
influencer replies to user-generated comments, the
more concerns that an influencer addresses. In con-
trast, positive comments are consistent with viewers’
positive impressions, so viewers of a video are not
actively seeking information to confirm their preexist-
ing beliefs.
The fourth explanation lies in parasocial interaction
(PSI), the tendency of individuals to regard media
characters as friends (Rubin, Perse, and Powell 1985).
It is likely that the increase in the frequency of influ-
encers’replies to comments would improve consumer
perception of the influencer and product via the psy-
chological mechanism of PSI. That is, the more fre-
quently an influencer interacts with followers/viewers
in the comment section, the more likely a
follower/viewer would be to perceive the influencer as
a friendly and trustworthy person.
A series of empirical inquiries investigated influen-
cer marketing from a PSI perspective. Yuan and Lou
(2020) examined antecedents of PSI and consumers’
interests in products promoted by influencers. Factors
such as the attractiveness of influencers, similarity to
influencers, procedural fairness, and fairness of inter-
actions with influencers were reported to be positively
correlated with the strength of PSI. Jin and Ryu
(2020) found that the type of photos moderated the
influence of PSI on consumers’intention to purchase
a product promoted by an Instagram influencer and
that content generator types also moderated the influ-
ence of PSI on consumers’judgment of source trust-
worthiness. Boerman (2020) investigated whether
brand disclosure and follower count affected consum-
ers’perception of PSI with influencers but did not
find a significant effect. Physical attractiveness and
attitude homophily were found to affect consumers’
perception of PSI and their attitude toward a brand
promoted by a YouTube influencer (Lee and Watkins
2016; Sokolova and Kefi 2020). Reinikainen et al.
(2020) discovered that comments posted by viewers of
an influencer-produced video moderated PSI’s influ-
ence on brand trust and purchase intention; however,
it was unclear whether the reply posted by an influen-
cer would further moderate the correlation. According
to the findings of the reviewed studies, it is evident
that PSI affects consumer perception of a product and
an influencer, and the interaction between influencers
and followers is likely to be the key that forges the
positive PSI. Based on the aforementioned reasonings,
the following hypothesis was proposed.
H
3
There will be a significant moderated mediation
effect. That is, the perceived trustworthiness of influ-
encer will mediate the correlation between comment
valence and product attitude; influencer-viewer inter-
action frequency will moderate the correlation
between comment valence and perceived influencer
trustworthiness and the correlation between comment
valence and product attitude.
The Significance of Studying Influencer–Follower
Interactions
Influencer marketing is enabled by and conducted on
digital social platforms such as Instagram, TikTok,
YouTube, and Facebook. The common thread among
social platforms is that they are designed to encourage
interpersonal interactions. On those platforms, inter-
personal interactions are often reflected in the form of
6 M. XIAO

conversations or exchanges of information between
message senders and receivers (i.e., comment and
reply). In the current research, the key constructs of
the investigation—comment valence and influencer
reply frequency—are closely related to the social
nature of influencer marketing. Many scholars and
industry experts believe that trust, cultivated by the
interaction between consumers (i.e., information
receivers) and influencers (i.e., information senders),
is the key to the success of influencer marketing
(Mettenheim and Wiedmann 2021). Thus, the investi-
gation of how the comment-and-reply interaction
between consumers and influencers affects consumer
perception is relevant to the research of influencer
marketing. Specifically, the contributions of the cur-
rent study can be discussed from the following
perspectives.
First, theory testing is context-dependent, and the
contexts of empirical studies are usually different. A
certain theoretical framework may have been
employed by several studies that examined a similar
set of variables in various contexts, but each study still
offers unique theoretical implications. For example,
the MAIN model is designed to explore and explain
heuristic information processing outcomes on digital
media. The model has been employed in studying
users of various digital media platforms such as news
media sites and Twitter (Xu 2013; Waddell and
Sundar 2017; Waddell 2018). The overall theoretical
framework and the key variables examined in those
studies are identical, but the findings or outcomes of
the studies are different because the contexts of the
studies vary. In the context of the current research,
the experience of watching YouTube videos produced
by an influencer is different from reading a news art-
icle or tweeting about a television program because
the goal of information consumption/dissemination
and the modality of information are different.
Therefore, the exact influence of factors, namely com-
ment valence and influencer reply frequency, on a
consumer’s perception of an influencer-produced
YouTube video should be ascertained by an empirical
investigation. The findings of the investigation should
offer unique theoretical contributions to our know-
ledge of the phenomenon.
Although the research design and method of the
current study can be applied to studying any
YouTube video, a key difference between regular
YouTube videos and influencer marketing videos
determines the uniqueness of the current research.
This key difference is whether a product or brand is a
focus of the video content. In influencer marketing
videos, the influencers need to serve as brand ambas-
sadors; hence, products or brands have to be discussed
or mentioned in the videos. However, the theme of a
regular YouTube video can vary greatly, and the host
of the video does not have the obligation to introduce
a brand or product in the video. One way for influ-
encers to maintain a healthy relationship with their
followers and to create a positive brand (or product)
image on behalf of their sponsors is by chatting and
interacting with their followers. Influencers need to
keep their marketing or branding duties in mind
when they are interacting with their fans. In contrast,
hosts of non-influencer videos on YouTube do not
have branding or marketing obligations when they
interact with their followers. Hence, an empirical
examination of influencer–viewer interaction in the
context of influencer marketing on YouTube would
generate unique findings that are different from what
has been unearthed by studies that examined
YouTube videos as a general topic.
Furthermore, the current research project has
sampled two different groups of audiences (i.e., college
students and general U.S. consumers). The method of
research has been replicated in three experiments (i.e.,
a pretest and two main studies), and the findings have
been validated by the results of the studies. Thus, this
research project is one of a few attempts to address
the so-called replication crisis faced by social scientists
(Shrout and Rodgers 2018). As of now, an empirical
research study in social science normally examines a
specific aspect of a topic or phenomenon. Each study
may examine a set of variables that are relevant to the
overarching topic or phenomenon, and the findings
improve our understanding of a phenomenon from
the perspective of a particular set of variables. The
accumulation of scientific evidence obtained from a
series of studies examining the same topic or phe-
nomenon gradually advances our understanding of
the topic so that, one day, our knowledge about the
topic would be revolutionized (Kuhn 1970). Thus, the
current research is part of an effort to advance our
understanding of influencer marketing by using a sys-
tematic and scientific research method. The findings
of the current research would also benefit future
scholars who investigate the topic of influencer mar-
keting by using meta-analysis as the research method.
The current research is different from other litera-
ture about eWOM and product reviews in that the
findings of the research would enrich our knowledge
of eWOM from a perspective of influencer–follower
interaction. The key difference between this study and
the other product review literature can be discussed
JOURNAL OF INTERACTIVE ADVERTISING 7

from three aspects. First, many empirical studies of
product reviews have examined the influence of
reviews without examining how the interaction
between the users/consumers and sellers/brands affects
consumer perception (e.g., Filieri, Raguseo, and Vitari
2021; Hong and Pittman 2020).
Second, the studies that have examined “replies”
investigated the issue from a perspective of a seller
but not from a perspective of an influencer (Le and
Ha 2021; Zhao, Jiang, and Su 2020; Wu et al. 2020).
Consumers may regard retailers, sellers, or brands as
entities or parties that have an overt financial interest
in the products they promote because sellers or
brands are the ones that directly sell the product. In
contrast, consumers may not regard influencers in the
same way. Instead, many consumers perceive influ-
encers as an independent or a more trustworthy voice
(Schouten, Janssen, and Verspaget 2020). Hence, the
level of perceived trustworthiness of a retailer and an
influencer must be different. The varying degree of
trust determines that the influencer–follower inter-
action is different from the seller–consumer inter-
action. Hence, a research study examining influencer–
follower interaction and its influence on consumers is
needed. The findings of such a study would broaden
the scope of literature about eWOM and product
reviews.
Third, the modality of information presentation on
e-commerce sites and YouTube is different.
Consumers are likely to evaluate user-generated
eWOM communication (e.g., product reviews, com-
ments) based on existing information presented on a
website. The existing information on e-commerce sites
is presented in various formats such as text-based
product descriptions, pictures of a product, and short
videos. On YouTube, the existing information is pre-
sented predominantly in the format of video.
According to Appiah (2006), the information pre-
sented in the video/audio format is considered more
vivid and richer than information presented in the
text/picture format. Thus, the way consumers process
existing information on an e-commerce site is differ-
ent from how they process such information on
YouTube, and comments and replies accompanying a
YouTube video may influence consumers in a differ-
ent way than do reviews on e-commerce sites.
Pretest
Method
A laboratory experiment was conducted at a midwest-
ern university in the United States. The experiment
had three conditions: negative comment condition,
positive comment condition, and no comment control
condition. Participants were randomly assigned to one
of the conditions. A total of 106 undergraduate stu-
dents were sampled. The average age of participants
was 20.64 (SD ¼3.90). The sample had 52 male partic-
ipants and 53 female participants. Most participants
were White (n¼60, 56.6%), and the rest of the partic-
ipants were Asian (n¼19, 17.9%), Hispanic (n¼12,
11.3%), Black or African American (n¼8, 7.5%), or
other (n¼7, 6.7%). Each participant received extra
course credits as an incentive.
Stimuli
A professionally produced product review video was
used as the experimental stimulus. A few seconds of
the original video were deleted because the influencer
talked about ways to follow him on social media. The
video was created by a moderately well-known
YouTube influencer (subscriber count ¼1.3 million),
who specializes in producing unboxing and product
review videos about consumer electronic products. In
the video used as the experimental stimulus, the influ-
encer reviewed a lesser-known product—Nanoleaf
Canvas lighting panels. The video demonstrated the
functions and features of the product such as how to
install lighting panels and how to use smartphones to
control the panels. Because the influencer introduced
himself as Mike in the video, a mock YouTube chan-
nel was created, and “Mike”was used as the name of
the channel host. The influencer’s number of followers
was not revealed on the mock YouTube channel or to
the participants at any point during the experiment.
The influencer is a middle-aged, White American.
In the video used for this study, the influencer
described and reviewed the product by using a voice-
over and thus did not reveal his face to the viewers.
Therefore, the influence of the appearance of an influ-
encer was controlled by using such a video. In gen-
eral, this experiment aims to emulate a scenario
similar to what a consumer may experience when the
consumer is researching a new or unfamiliar product
by watching a YouTube video.
Regarding the stimuli of user-generated comments,
comments posted by actual viewers of the video on
the influencer’s channel as well as reviews of the
product posted on the brand’s online store were col-
lected. The collected negative comments/reviews were
mainly attacking the product from the aspects of price
and design, while the positive comments/reviews were
mainly praising the design of the product. Six negative
comments and six positive comments were created
8 M. XIAO

based on the collected comments/reviews. Six Google
accounts were created to post the comments on the
mock channel. The usernames of the mock accounts
for posting comments were combinations of generic
English names and random usernames selected from
the influencer’s channel. The usernames were ran-
domly associated with the comments. The position of
each comment was randomly assigned on the web-
page. A manipulation check question was placed in
the survey questionnaire [What do you think about
user-generated comments? They are (1) very negative
…(7) very positive]. Examples of stimuli were
attached.
Comment valence was manipulated as an inde-
pendent variable by using the stimuli described in the
above section (n
negative comment
¼33, n
positive comment
¼37, n
control condition
¼36). Product attitude was
measured as a dependent variable. In branded
YouTube videos, such as some unboxing and product
review videos, product-related information is what
attracts viewership. To brands and marketers, a clear
understanding of how the presentation of a product
in a video influences product attitude is the key to
helping brands and marketers fine-tune their
YouTube marketing campaigns. Thus, measuring
product attitude is not only meaningful to academic
scholars in assessing consumer perception of video
content but also useful to brands and marketers in
improving the quality of their marketing endeavors.
The construct of product attitude was measured by
five items on a 7-point semantic differential scale
[e.g., (1) unappealing to (7) appealing] adapted from
Spears and Singh (2004)(a¼.945).
Perceived trustworthiness of the influencer was
measured as the other dependent variable. The percep-
tion of information source trustworthiness is positively
related to perceived information quality and credibility
(McGinnies and Ward 1980). The apparent trust-
worthiness of an influencer is important to a brand
because it affects the persuasiveness of product-related
information delivered via a video. Thus, brands and
influencers need to understand how user-generated
comments may affect one’s perceived trustworthiness
of an influencer so that the brand or influencer can act
accordingly to either mitigate the negative impact or
promote the positive influence. The construct was
measured by five items on a 7-point semantic differen-
tial scale [e.g., (1) undependable to (7) dependable]
adapted from Ohanian (1990)(a¼.921).
A person’s familiarity with the product, brand,
influencer, and video were controlled in the experi-
ment. The scale items that measured these variables
were adapted from the study of Flynn and Goldsmith
(1999). Four items, such as “I’m familiar with the
product/brand/influencer/video”and “I know the pro-
duct/brand/influencer/video very well,”were measured
on a 7-point scale from (1) strongly disagree to (7)
strongly agree (a
product familiarity
¼.882, a
brand familiarity
¼.801, a
influencer familiarity
¼.894, a
video familiarity
¼.847).
Procedure
Participants were instructed to fill out a sign-in form
upon their arrival. The researchers then assigned them
to a computer. The computers were operating on Mac
OS 10.14 system. Each computer was equipped with
headphones of the same brand and model. The
experimental stimuli were displayed on the Safari web
browser. The browsing history was cleared after each
session of the experiment. Participants were instructed
to watch the video first and then to read related infor-
mation (i.e., comments) presented on the webpage.
Participants were asked to complete a questionnaire in
Qualtrics after they finished watching and reading the
experiment stimuli. Participants were thanked and dis-
missed after the completion of the questionnaire.
Results
The manipulation of comment valence was examined
in an independent samples ttest, and the results
revealed that the manipulation was successful, t(59.36)
¼63.90, p<.001. One-way analysis of variance
(ANCOVA) tests were performed to examine the
potential association between comment valence and
the dependent variables. The results revealed that
comment valence significantly influenced product atti-
tude, F(2, 99) ¼29.97, p<.001, partial g
2
¼.381,
and perceived trustworthiness of influencer, F(2, 99)
¼5.84, p¼.004, partial g
2
¼.112. Specifically, par-
ticipants who were assigned to the negative comment
condition had a significantly more negative attitude
toward the product (M¼4.87, SE ¼.17) than those
that were assigned to the positive comment (M¼6.46,
SE ¼.16, p<.001) or the control condition
(M¼6.47, SE ¼.16, p<.001). However, the influ-
ence of positive comments was not statistically differ-
ent from that of the no comment control condition.
Similarly, participants who were assigned to the nega-
tive comment condition considered the influencer less
trustworthy (M¼5.10, SE ¼.18) than those who were
assigned to the positive comment (M¼5.86, SE ¼.17,
p¼.011) or the control condition (M¼5.86, SE ¼
.17, p¼.010). However, the influence of positive
JOURNAL OF INTERACTIVE ADVERTISING 9

comments on perceived trustworthiness of influencer
was not significantly different from that of the no
comment condition.
To examine the potential mediation effect of trust-
worthiness on product attitude, two separate interaction
terms for comment valence were created. Specifically, the
first interaction term compares the difference between
the negative comment condition and the no comment
control condition. The second interaction term compares
the difference between the positive comment condition
and the control condition. The PROCESS macro in SPSS
was employed to analyze the data.
When examining the first interaction term (nega-
tive vs. control), we found that comment valence (b
¼.824, p<.001) significantly influenced perceived
trustworthiness of influencer. Moreover, comment
valence (b¼1.475, p<.001) significantly influ-
enced product attitude. However, the mediation index
suggested that the proposed mediation effect was not
significant (.1369; LLCI: .4024 to ULCI: .0481).
Please note that the 95% confidence interval (CI) level
was used in the current study.
When examining the second interaction term (posi-
tive vs. control), we found that comment valence did
not influence perceived trustworthiness of influencer
(b¼.015, p¼.893) or product attitude (b¼.0004,
p¼.995). The mediation effect was not significant
(.0022; LLCI: .0362 to ULCI: .0428).
Study 1
Method
Although part of the findings unearthed in the pretest
is significant, the sample of the study has limitations.
That is, undergraduate students are younger than
general consumers; thus, they are more likely to be
tech-savvy or more receptive to innovation than the
general public (Lee, Atkinson, and Sung 2020). Hence,
a more representative sample is needed to validate the
findings of the pretest and to improve the generaliz-
ability of the empirical discoveries. Study 1 was con-
ducted by sampling MTurk users.
The experiment had three conditions: negative com-
ment condition (n¼38), positive comment condition
(n¼40), and no comment control condition (n¼41).
The screening criteria on MTurk were set at HIT
approval rate 99% and number of HITs approved
1,000, and the location is the U.S. or Canada. A total
of 150 MTurk users were sampled. Thirty-one
responses were not included in the data analysis
because the participants either did not provide a rea-
sonable answer to the manipulation check questions of
comment valence (e.g., assigned to the positive com-
ment condition but rated comments negatively) or did
not provide a meaningful response (e.g., “ABCDE”)to
the attention check question that asked participants to
leave a hypothetical comment for the video. The final
sample size was 119. The average age of participants
was 36 (SD ¼11.52, ranging from 19 to 70). The sam-
ple had 70 male participants and 48 female partici-
pants. Most participants were White (n¼87, 73.1%).
The ethnicities of the rest of the participants were
Asian (n¼12, 10.1%), Hispanic (n¼10, 8.4%), and
Black or African American (n¼9, 7.6%). Each partici-
pant received $1.50 for participation.
The experiment stimulus was the same video used
in the pretest. The same measurement scales used in
the pretest were utilized again. Product attitude (a¼
.983) and perceived trustworthiness of influencer (a¼
.962) were the dependent variables. A person’s famil-
iarity with the product, brand, influencer, and video
were controlled for in the experiment (a
product familiarity
¼.851, a
brand familiarity
¼.863, a
influencer familiarity
¼
.876, a
video familiarity
¼.850).
Procedure
In the experiment conducted on MTurk, participants
were randomly assigned a hyperlink that directed
them to visit an experiment condition. They were
instructed to watch the video first and then to read
comments posted on the webpage. Participants were
informed to return to the questionnaire after watching
and reading the experiment stimuli.
Results
The manipulation of comment valence was successful,
t(76) ¼33.10, p<.001. Participants who were
assigned to the negative comment condition consid-
ered comments negatively (M¼1.55, SD ¼.76),
whereas those who were assigned to the positive com-
ment condition considered the comments positively
(M¼6.70, SD ¼.61). In addition to the use of
manipulation checks, most participants were not
familiar with the product (M¼2.27, SD ¼1.47), brand
(M¼2.26, SD ¼1.53), influencer (M¼1.91,
SD ¼1.41), or video (M¼2.04, SD ¼1.42).
1
A one-way ANCOVA was conducted, and the
result revealed that comment valence significantly
influenced product attitude, F(2, 112) ¼47.498, p<
.001, partial g
2
¼.459. A post-hoc analysis was per-
formed with a Bonferroni adjustment. The adjusted
marginal means of the Bonferroni tests were exam-
ined. Participants who were assigned to the negative
10 M. XIAO

comment condition had a significantly more negative
attitude toward the product (M¼4.357, SE ¼.176)
than those who were assigned to the positive com-
ment (M¼6.443, SE ¼.172, p<.001) or the no
comment condition (M¼6.466, SE ¼.166, p<.001).
However, the influence of positive comments was not
significantly different from that of the no comment
control condition. Thus, hypothesis 1 was supported.
2
To test the influence of comment valence on per-
ceived trustworthiness of influencer, another one-way
ANCOVA was conducted. The result revealed a sig-
nificant impact of comment valence on perceived
trustworthiness of influencer, F(2, 112) ¼18.564, p<
.001, partial g
2
¼.249. A post hoc analysis was per-
formed with a Bonferroni adjustment. Participants
who were assigned to the negative comment condition
considered the influencer less trustworthy (M¼5.151,
SE ¼.151) than those who were assigned to the posi-
tive comment (M¼6.382, SE ¼.147, p<.001) or the
control condition (M¼6.155, SE ¼.143, p<.001).
However, the influence of positive comments on per-
ceived trustworthiness of influencer was not signifi-
cantly different from that of the control condition.
Thus, hypothesis 2 was supported. The results of the
ANCOVA tests are displayed in Tables 1 and 2.
Similar to what has been conducted in the pretest,
the potential mediation effect of perceived influencer
trustworthiness on product attitude was examined.
When examining the first interaction term (negative vs.
control) for comment valence, we found that comment
valence (b¼.506, p¼.016) significantly influenced per-
ceived trustworthiness of influencer. Moreover,
comment valence (b¼1.109, p<.001) significantly
influenced product attitude. Furthermore, the mediation
index suggested that the proposed mediation effect was
significant (.3161; LLCI: .0561 to ULCI: .6840).
When examining the second interaction term (posi-
tive vs. control), we found that comment valence did
not influence perceived trustworthiness of influencer
(b¼.0604 ¼.737) or influence product attitude (b
¼.095, p¼.534). The mediation effect was not sig-
nificant (.0242; LLCI: .1606 to ULCI: .1273). The
mediation model and path coefficients are displayed
in Figure 1.
Study 2
Method
Study 2 was conducted on MTurk to examine the effi-
cacy of using frequent influencer–viewer interaction
as a strategy to alleviate the influence of negative
comments on consumers. This study was a 2 (negative
vs. positive comments) 3 (influencer–viewer
interaction: low-frequency vs. mid-frequency vs. high-
frequency) between-subjects experiment. The screen-
ing criteria on MTurk were set at HIT approval rate
99% and number of HITs approved 1,000, and the
location is the U.S. or Canada. An additional filter
was added to prevent individuals who had participated
in Study 1 from participating in the current study. A
total of 240 MTurk users were sampled. Thirty-nine
responses were not included in the data analysis
because the participants either did not provide a rea-
sonable answer to the manipulation check questions
or did not provide a meaningful response to the atten-
tion check question that asked participants to write a
Table 1. Means, standard deviations, adjusted means, and
standard errors for product attitude measured in study 1.
Experiment conditions
Negative comments No comment Positive comments
M4.305 6.478 6.480
(SD) (1.676) (.583) (.616)
M
adj
4.357 6.466 6.443
(SE) (.176) (.166) (.172)
Note: M: Means; SD: Standard Deviations; SE: Standard Errors; M
adj
:
Adjusted Means.
Table 2. Means, standard deviations, adjusted means, and
standard errors for perceived trustworthiness of influencer
measured in study 1.
Experiment conditions
Negative comments No comment Positive comments
M5.194 6.136 6.360
(SD) (1.247) (.810) (.623)
M
adj
5.151 6.155 6.382
(SE) (.151) (.143) (.147)
Note: M: Means; SD: Standard Deviations; SE: Standard Errors; M
adj
:
Adjusted Means.
.625*** (.399***)
.506* (–.0604)
Comment
Valence
Perceived
Influencer
Trustworthiness
Product
Attitude
1.109*** (–.095)
Figure 1. The mediation model and path coefficients (study
1). Path coefficients outside the parentheses represent the
coefficients obtained from analyzing the first interaction term
(negative comments vs. no-comment control) and the medi-
ation index was .3161 (LLCI: .0561 to ULCI: .6840). Path coeffi-
cients inside the parentheses represent the coefficients
obtained from analyzing the second interaction term (positive
comments vs. control), and the mediation index was –.0242
(LLCI: –.1606 to ULCI: .1273). p<.05; p<.001.
JOURNAL OF INTERACTIVE ADVERTISING 11

hypothetical comment. The final sample size was 201.
The average age of participants was 33 (SD ¼11.84,
ranging from 18 to 68). The sample had 108 male
participants and 89 female participants. Most partici-
pants were White (n¼143, 71.1%). The ethnicities of
the rest of the participants were Asian (n¼23,
11.4%), Black or African American (n¼16, 8%),
Hispanic (n¼11, 5.5%), and other (n¼8, 4%). Each
participant received $2.00 as an incentive.
In addition to the video and user-generated com-
ments, an influencer’s replies to user-generated com-
ments were manipulated. The replies were adapted
from the actual replies posted on the influencer’s
YouTube channel. The collected replies were revised
and reorganized based on the content of user-gener-
ated comments utilized in the previous experiment.
Overall, the replies to negative comments were manip-
ulated as counterarguments to the comments, while
the replies to positive comments were manipulated as
affirmations to the comments. Although the valence
of replies differs, the content of negative and positive
replies is similar. In fact, the points discussed in
replies to negative comments and replies to positive
comments are identical because positive replies were
created based on replies to negative comments. Thus,
the confounding effects would be minimized by such
a way of manipulation. Please see the Appendix for
screenshots of experiment stimuli.
Comment valence (n
negative comment condition
¼102,
n
positive comment condition
¼99) was manipulated by
using the “comment”stimuli described in the previous
experiment. In the low interaction frequency condition
(n¼68), no reply was provided by the influencer. In
the mid interaction frequency condition (n¼65), two
replies were posted by the influencer with the positions
of the replies randomly selected. In the high interaction
frequency condition (n¼68), the influencer replied to
all six comments. The manipulation check question
[“This YouTube influencer (1) did not interact with
viewers at all …(7) interacted with viewers all the
time”] was placed in the survey instrument. Regarding
the dependent variables, product attitude (a¼.935)
and perceived trustworthiness of influencer (a¼.926)
were measured. The dependent variables were meas-
ured on the same scales used in Study 1.
Results
An independent samples ttest was conducted to
examine the manipulation of comment valence, and
the results revealed that the manipulation was success-
ful, t(199) ¼56.982, p<.001. Participants assigned
to the negative comment conditions considered the
comments negatively (M¼1.38, SD ¼.661), while
participants in the positive comment conditions con-
sidered the comments positively (M¼6.64, SD ¼
.646). A one-way analysis of variance (ANOVA) was
conducted to examine the manipulation of influencer–
viewer interaction frequency, and the results revealed
that the manipulation was successful, F(2, 198) ¼
1,105.619, p<.001. Participants considered the fre-
quency of interaction between the influencer and
viewers as low (M¼1.04, SD ¼.207), medium
(M¼3.92, SD ¼.941), and high (M¼6.71, SD ¼
.754) in the low-frequency, mid-frequency, and high-
frequency conditions, respectively. In general, partici-
pants were not familiar with the product (M¼2.21,
SD ¼1.491), brand (M¼2.15, SD ¼1.422), influencer
(M¼1.67, SD ¼1.052), and video (M¼1.79,
SD ¼1.198). Given the low familiarity scores reported
by participants, the potential covariates were success-
fully controlled by the stimuli. Thus, they were not
included in the data analysis.
To further delineate the difference across condi-
tions in the moderated mediation model, two separate
interaction terms for reply frequency were created.
Specifically, the first interaction term compares the
difference between the high-frequency and no-reply
conditions. The second interaction term compares the
difference between the mid-frequency and no-reply
conditions. The PROCESS macro in SPSS was
employed to analyze the moderated mediation model.
When examining the first interaction term (high-
frequency vs. no-reply), the results suggest that com-
ment valence (b¼1.111, p<.001), frequency of reply
(b¼.998, p<.001), and the interaction effect
between frequency of reply and comment valence (b
¼.932, p<.001) significantly influenced perceived
trustworthiness of influencer. Moreover, comment
valence (b¼1.09, p<.001), frequency of reply (b¼
.748, p<.001), perceived trustworthiness of influen-
cer (b¼.373, p<.001), and the interaction effect
between comment valence and frequency of reply (b
¼.631, p¼.028) also significantly influenced prod-
uct attitude. Furthermore, the omnibus moderated
mediation index suggested that the proposed moder-
ated mediation effect was significant (.3485; LLCI:
.6830 to ULCI: .1110).
When examining the second interaction term (mid-
frequency vs. no-reply), the results suggest that com-
ment valence (b¼1.112, p<.001), frequency of reply
(b¼.493, p¼.014), and the interaction effect
between frequency of reply and comment valence (b
¼.826, p¼.004) significantly influenced perceived
12 M. XIAO

trustworthiness of influencer. Moreover, comment
valence (b¼1.083, p<.001), frequency of reply (b¼
.569, p¼.006), perceived trustworthiness of influen-
cer (b¼.381, p<.001), and the interaction effect
between comment valence and frequency of reply (b
¼.688, p¼.020) also significantly influenced prod-
uct attitude. Furthermore, the omnibus moderated
mediation index suggested that the proposed moder-
ated mediation effect was significant (.3144; LLCI:
.6443 to ULCI: .0764). Thus, hypothesis 3 was
supported. The moderated mediation model and path
coefficients are displayed in Figure 2.
A one-way ANOVA was performed to examine the
main effect of comment valence on perceived trust-
worthiness of influencer. The main effect of comment
valence, F(1, 199) ¼20.513, p<.001, partial g
2
¼
.093, was significant. Participants who read negative
comments (M¼5.567, SD ¼.861) perceived the influ-
encer as less trustworthy than those who read positive
comments (M¼6.091, SD ¼.776).
Another ANOVA was performed for product atti-
tude. The main effect of comment valence, F(1, 199)
¼47.784, p<.001, partial g
2
¼.194, was significant.
Participants who read negative comments (M¼5.594,
SD ¼1.006) perceived the product less favorably than
those who read positive comments (M¼6.438, SD ¼
.962). The results of the ANOVA tests are displayed
in Table 3.
3
General Discussion
The findings of the experiments confirm the overpow-
ering impact of negative comments on product
attitude and perceived trustworthiness of influencers.
The findings also suggest that frequent influencer–
viewer interaction can be a remedy to soothe the
damage inflicted by negative comments. The proposed
hypotheses are mostly supported by the data.
Theoretical Implications
The pretest and study 1 examined YouTube influencer
marketing from an angle of user-generated comments,
psychological heuristics, and the negativity bias that
extant studies of influencer marketing have over-
looked (Lou and Yuan 2019; De Veirman, Cauberghe,
and Hudders 2017; Jin and Ryu 2020). Unlike some of
the existing studies in which product review valence
was found to be influential in both negative and posi-
tive directions (e.g., Chen and Lurie 2013; Hong and
Pittman 2020; Lee, Atkinson, and Sung 2020), the cur-
rent research has unearthed a disproportionate influ-
ence of negative comments on consumers. The
findings of the study offer two distinctive implica-
tions. First, the influence of information cues on con-
sumers varies in different media consumption and
heuristic processing contexts. Second, viewers of a
product-related YouTube video perceive a product
and an influencer based on the opinion of other view-
ers when the general opinion is negative.
From a bandwagon heuristic perspective, a group
of positive comments on YouTube did not activate a
positive bandwagon heuristic given that the measure-
ment results of the dependent variables (i.e., product
attitude and perceived trustworthiness of influencer)
in the positive comment condition were identical to
the results obtained from the control condition. In
contrast, negative comments exerted a strong influ-
ence on product attitude and perceived trustworthi-
ness of influencer. The findings of the current
research are different from what has been reported in
extant research regarding the influence of the positive
bandwagon heuristic (Fu and Sim 2011; Sundar,
Oeldorf-Hirsch, and Xu 2008), but the findings are
consistent with extant research about the negativity
bias and negative bandwagon heuristic (Rozin and
.373 (.381)
–.631 (–.688)
–.932 (–.826)
1.111 (1.112)
Comment
Valence
Perceived
Influencer
Trustworthiness
Product
Attitude
Interaction
Frequency
1.090 (1.083)
Figure 2. The moderated mediation model and path coeffi-
cients (study 2). Path coefficients outside the parentheses rep-
resent the coefficients obtained from analyzing the first
interaction term (high reply frequency vs. no-reply), and the
moderated mediation index was –.3485 (LLCI: –.6830 to ULCI:
–.1110). Path coefficients inside the parentheses represent the
coefficients obtained from analyzing the second interaction
term (mid reply frequency vs. no-reply), and the moderated
mediation index was –.3144 (LLCI: –.6443 to ULCI: –.0764). All
path coefficients were significant (p<.05).
Table 3. Means and standard deviations for the influence of
comment valence on product attitude and perceived trust-
worthiness of influencer measured in study 2.
Experiment conditions
Negative comments Positive comments
Product attitude, M(SD) 5.594 6.438
(1.006) (.962)
Trustworthiness, M(SD) 5.567 6.091
(.861) (.776)
Note: M: Means; SD: Standard Deviations.
JOURNAL OF INTERACTIVE ADVERTISING 13

Royzman 2001; Waddell and Sundar 2017), indicating
the ubiquitous presence of the negativity bias phe-
nomenon in our lives. The strong impact of negative
comments on individuals can be explained from two
perspectives. First, individuals pay more attention to
negative events than to positive events because the
occurrence of negative events is rarer than that of
positive events (Lewicka, Czapinski, and Peeters
1992). Second, the loss aversion mentality leads indi-
viduals to be attentive when they encounter the nega-
tivity (Tversky and Kahneman 1991). In other words,
individuals’attentiveness to negative events leads
them to better retain negative information that influ-
ences their perception of a certain issue.
The increase in the frequency of influencer–viewer
interaction remediates consumers’overall perception
of an influencer and a product promoted by the influ-
encer when consumers are exposed to negative user-
generated comments. The findings implicate that the
increase in the interaction frequency cues consumers
to form an impression of the influencer as a reliable
and trustworthy person who is willing to communi-
cate with fans and followers. This positive image of an
influencer is likely to mitigate the negative impact
generated by negative comments. The findings also
implicate that an influencer’s replies to user-generated
comments offer information to potentially reconcile
concerns or dissatisfactions expressed by consumers
in negative comments. The more frequent the replies,
the more concerns an influencer may address.
The theoretical contribution of study 2 can be dis-
cussed from the contingency heuristic perspective.
When watching a product-related video on YouTube,
viewers can obtain adequate information about a
product from the video. However, the video may
reveal little about the influencer, and viewers may not
be familiar with the influencer. Thus, viewers must
utilize heuristics, such as the contingency heuristic, to
help them evaluate the credibility of the information
source. The more frequent the influencer–viewer
interactions, the more information an influencer
seems to offer in the discourse, and thus the more
contingent the information exchange process will be
perceived by viewers (Sundar et al. 2016). A clear con-
tingent perception leads viewers to perceive the inter-
action between the influencer and followers as unique,
timely, and reliable (Sundar 2008). Hence, the per-
ceived trustworthiness of an influencer improves as
the frequency of interaction between the influencer
and viewers increases.
Moreover, the moderating impact of influencer–
viewer interaction frequency on consumer perception
can be discussed from a parasocial interaction per-
spective. The increase in the level of reply frequency
is likely to cultivate a positive image of an influencer
as a caring individual in the heart of viewers. As a
result, consumers’perception of their parasocial rela-
tionship with the influencer is likely to be enhanced
and consumer perceptions of influencer trustworthi-
ness and product attitude may improve. This phe-
nomenon is even more evident when the valence of
user-generated comments is negative.
From a confirmation bias perspective, the findings
of study 2 implicate that most viewers of the video
had a positive first impression of the influencer and
the product promoted by the influencer. However, the
negative comments underneath the video contradicted
viewers’preexisting positive impressions. Thus, when
viewers read the influencer’s replies, which were
manipulated as counterarguments to negative com-
ments, viewers used the replies as additional informa-
tion to help them reaffirm their preexisting beliefs
(Klayman 1995; Metzger and Flanagin 2013). In con-
trast, when viewers encountered positive comments,
the positive comments and the influencer’s replies to
these comments did not significantly affect viewers’
opinions because the information was consistent with
viewers’established beliefs (Nickerson 1998).
However, it should be noted that the conceptual foun-
dations (e.g., contingency heuristic, confirmation bias,
and parasocial interaction) have been used as the
explanatory mechanisms in the current study that the
constructs were not directly measured. Future research
projects should measure the constructs in order to
obtain more conclusive evidence to better explain the
cause of the phenomenon.
Furthermore, a ceiling effect may exist when con-
sumers read positive user-generated comments while
witnessing an increase in reply frequency. It is likely
that the positive opinion shaped by positive comments
is so strong that even the most frequent interaction
initiated by the influencer cannot further improve
such a positive opinion any further. Another explan-
ation is that the 7-point measurement scale may have
limited scale points that the variance or improvement
in product attitude and perceived trustworthiness of
influencer cannot be discerned or detected by the cur-
rent instrument. Future studies may wish to employ a
scale that contains more scale points to capture the
variance between conditions.
The findings of three experiments suggest that the
mediating influence of perceived influencer trust-
worthiness is context-dependent. The pretest sampled
young college students. However, the mediation effect
14 M. XIAO

of perceived influencer trustworthiness was not signifi-
cant in the pretest. In contrast, study 1 sampled general
U.S. consumers on MTurk and discovered a significant
mediation effect exerted by perceived influencer trust-
worthiness on the correlation between negative com-
ments and product attitude. Similarly, study 2
unearthed a significant moderated mediation effect that
involves comment valence, reply frequency, and per-
ceived influencer trustworthiness. The aforementioned
findings offer unique theoretical implications.
First, multiple follow-up experiments were con-
ducted in this research project to replicate the results
obtained from an earlier investigation in different
populations. Few empirical investigations on influen-
cer marketing have been designed and conducted in
this way to address the replication issues. The chang-
ing mediation effect of perceived influencer trust-
worthiness across three experiments indicates the
necessity for future scholars to conduct multiple stud-
ies within a research project to replicate and validate
the investigation results.
Second, academic research studies that utilized
experiments as the research method tend to sample
college students. Some scholars argue that the sample
of college students offers valid results, while others
question the representativeness of such a sample. The
difference in the mediation effect across three experi-
ments implicates the difference in the mentality of con-
sumers from different demographic groups. Thus, the
findings suggest that it is imperative for advertising or
marketing scholars to sample a diverse group, or per-
haps multiple groups, of consumers in order to gain an
accurate understanding of consumer psychology.
Third, the significant moderated mediation effect
unearthed in study 2 indicates a complicated associ-
ation among factors that influence how consumers
perceive a product promoted by an influencer.
Existing studies have examined perceived trustworthi-
ness as a predictor of perceived attitude/credibility
(e.g., Schouten, Janssen, and Verspaget 2020; Xiao,
Wang, and Chan-Olmsted 2018) or as a construct
influenced by user-generated comments (e.g.,
Reinikainen et al. 2020), however, without examining
the impact of influencers on the mediator (i.e., per-
ceived trustworthiness) through the lens of comment-
and-reply interaction. The significant interaction
effects between comment valence and reply frequency
reveal that the dynamic of comment valence and its
influence on perceived trustworthiness of influencer
can be altered by how an influencer interacts with fol-
lowers/viewers. These factors ultimately shape a con-
sumer’s perception of a product.
In summary, this study found support for the sig-
nificant influence of comment valence and influencer–
viewer interaction on perceived trustworthiness of
influencer, and the findings revealed complex yet
coherent conceptual connections among the afore-
mentioned variables. The findings of the experiments
would be beneficial for future scholars to conduct a
meta-analysis to advance our understanding of influ-
encer marketing. Moreover, the findings of the studies
have the potential to enrich the body of the literature
related to the MAIN model and negativity bias (e.g.,
Sundar, Oeldorf-Hirsch, and Xu 2008; Waddell 2018)
and to expand the boundary of the knowledge about
the interplay between information cues and heuristic
processing. The overall contributions of the current
study can be discussed from three perspectives. First,
source credibility as a construct has been examined,
but not from an angle of heuristic processing as a
result of the bandwagon effect in the context of influ-
encer marketing. This study has examined this
important issue from a theoretical angle of informa-
tion cues and heuristics, expanding and extending the
scope of investigation of influencer-related research.
Second, this research project is one of a few endeavors
that investigate the interaction between influencer and
followers/audiences on YouTube in controlled experi-
ments. Thus, the findings of the study would assist us
in discerning the pattern and tendency of how an
individual interacts with an influencer when watching
branded videos on YouTube or other video-centric
platforms. Third, this study examined the negativity
bias phenomenon under the theoretical framework of
the MAIN model. Specifically, the study examined
how information cues, heuristics, and the negativity
bias affect the way an individual processes informa-
tion. Therefore, the findings would help us have a
more comprehensive view regarding the theoretical
connection between constructs, such as the negativity
bias and heuristic processing, and the way they deter-
mine the outcomes of online interpersonal
interactions.
Managerial Implications
The findings of the current research offer practical
implications that will help influencers, brands, and
advertisers conduct influencer marketing on YouTube
and other social media. The findings demonstrate the
effectiveness of using influencer–viewer interaction as
a way to maintain a positive image of an influencer. A
positive image of an influencer is the key to sustaining
a healthy relationship between the influencers and the
JOURNAL OF INTERACTIVE ADVERTISING 15

followers. Moreover, because influencers’replies to
positive comments do not significantly affect con-
sumer attitude, an influencer can simply “like”posi-
tive comments instead of spending extra effort
replying to these comments.
Although effective, many influencers do not reply
(or do not reply frequently enough) to negative com-
ments for various reasons. Thus, brands and market-
ers need to ensure that the influencer is actively
replying to consumers’concerns. One way for brands
to ensure that the influencers will perform their duty
is by specifying the requirement in the contract or
during the negotiation of the contract. For example, if
an influencer fails to reply to a certain number of
negative comments, the brands will have the right to
either terminate the collaboration or deduct the
amount of compensation for an influencer. Similarly,
a brand or marketer can offer influencers bonuses or
incentives to encourage them to reply to negative
comments as frequently as possible.
Other than influencers, brands and companies can
be the parties that reply to user-generated comments.
To maintain an ideal frequency of brand–customer
interactions, a brand or company may employ cus-
tomer service specialists whose job duties are to inter-
act with customers and to reply to customer-generated
comments on various social platforms. Specifically, the
brand—or the customer service specialists hired by the
brand—should apologize to unsatisfied customers and
offer constructive solutions to address the concerns
expressed in negative comments because doing so will
help the brand retain loyal customers and build a posi-
tive image. On the other hand, if the content of the
negative comments is malicious or baseless, such as the
ones posted by Internet trolls, a brand or a company
should provide evidence to counter the negative argu-
ments posted in the comments in order to halt the
spread of false information. Finally, a brand should
acknowledge customers’support expressed in positive
comments by liking the positive comments or by send-
ing a brief “thank you”message.
Limitations and Future Directions
First, the study did not examine how a mixed com-
ment condition (e.g., half negative comments and half
positive comments) would influence product attitude
and perceived trustworthiness of influencer. Future
studies should investigate how mixed comments
would affect consumers, so scholars can better discern
the phenomenon of negativity bias in the context of
influencer marketing. Second, a video stimulus was
used in three studies, whereas future studies should
utilize more than one video as experiment stimuli to
improve the generalizability of findings. Third, the
manipulation of experiment stimuli in study 2 did not
isolate the impact of the reasonings of the influencer’s
replies on participants. Participants of the study may
have paid attention to the content of the reply, or
they may have not read the replies but simply counted
the number of replies and formed the impression
based on the reply count. Future studies should differ-
entiate the two effects, so the impacts of the number
of replies and the content of replies can be isolated
and identified. Fourth, the bandwagon heuristic, con-
tingent heuristic, confirmation bias, and parasocial
interaction were employed to help explain phenomena
unearthed by the current research; however, the four
constructs were not measured. Future studies should
measure these constructs and analyze the data based
on the approach suggested by researchers, such as
Bellur and Sundar (2014), in order to obtain more
conclusive evidence to explain the cause of the phe-
nomenon. Fifth, this research project focused on eval-
uating how the frequency of influencer–viewer
interaction may influence product attitude and per-
ceived trustworthiness of influencer, while the actual
strategy (or method) of argumentation was not tested.
Future studies may focus on examining a specific
strategy or method of argumentation that an influen-
cer can employ to address different comments.
Finally, this research study utilized a lesser-known
product as a part of the stimuli. In reality, consumer
familiarity with a product or the experience of using a
product may be a key factor that determines con-
sumer perception. We, the consumers, have such a
tendency to value direct, personal experience as a
more reliable way of learning information than learn-
ing information vicariously through sources such as
comments or reviews. If a consumer has a positive
experience using a product, the consumer is likely to
have a positive attitude toward the product even
though the comments or reviews are mostly negative.
Future studies should investigate how product famil-
iarity affects consumer perception by using both well-
known and lesser-known products as the stimuli.
Notes
1. The results of the ANOVA tests in study 1 when
familiarity constructs were not controlled for and when
responses that contain misclassified comments were
included.
Trustworthiness of influencer: F(2, 131) ¼4.394, p<
.05, g
2
¼.055. The difference between the negative
16 M. XIAO

comment group (M¼5.498, SD ¼1.22) and the no
comment control group (M¼6.00, SD ¼.851) as well
as the difference between negative comments group and
positive comment group (M¼6.016, SD ¼.919) were
significant (p<.05).
Product attitude: F(2, 131) ¼21.480, p<.001, g
2
¼
.221. The difference between negative comment group
(M¼4.875, SD ¼1.77) and the no comment control
group (M¼6.30, SD ¼.761) as well as the difference
between negative comments group and positive
comment group (M¼6.204, SD ¼1.385) were
significant (p<.001).
2. Additional one-way ANCOVA tests were conducted for
study 1, in which gender and age were controlled for.
Gender, F(1, 110) ¼.643, p¼.424, and age, F(1, 110)
¼1.856, p¼.176, were not significant predictors of
perceived influencer trustworthiness. Similarly, gender,
F(1, 110) ¼.017, p¼.898, and age, F(1, 110) ¼.286, p
¼.594, did not significantly influence product attitude.
Hence, two factors did not affect the influence of
comment valence on the two dependent variables.
3. Additional two-way ANCOVA tests were conducted for
study 2 in which gender and age were controlled for. In
the equation that examined the comment valence’s and
interaction frequency’s influence on the dependent
variables, gender, F(1, 192) ¼.186, p¼.667, and age,
F(1, 192) ¼2.315, p¼.130, were not significant
predictors of perceived influencer trustworthiness.
Similarly, gender, F(1, 192) ¼.013, p¼.910, and age,
F(1, 192) ¼.080, p¼.777, did not significantly
influence product attitude.
ORCID
Min Xiao http://orcid.org/0000-0001-5367-4322
References
Appiah, O. 2006.“Rich Media, Poor Media: The Impact of
Audio/Video Vs. Text/Picture Testimonial Ads on
Browsers’Evaluations of Commercial Web Sites and
Online Products.”Journal of Current Issues & Research in
Advertising 28 (1): 73–86. doi:10.1080/10641734.2006.
10505192.
Boerman, S. C. 2020.“The effects of the Standardized
Instagram Disclosure for Micro- and Meso- Influencers.”
Computers in Human Behavior 103: 199–207. doi:10.
1016/j.chb.2019.09.015.
Bellur, Saraswathi, and S. Shyam Sundar. 2014.“How Can
We Tell When a Heuristic Has Been Used? Design and
Analysis Strategies for Capturing the Operation of
Heuristics.”Communication Methods and Measures 8 (2):
116–137. doi:10.1080/19312458.2014.903390.
Boerman, S. C., and E. Reijmersdal. 2019.“Disclosing
Influencer Marketing on YouTube to Children: The
Moderating Role of Para-Social Relationship.”Frontiers
in Psychology 10: 3042. doi:10.3389/fpsyg.2019.03042.
Chaiken, S. 1980.“Heuristic Versus Systematic Information
Processing and the Use of Source versus Message Cues in
Persuasion.”Journal of Personality and Social Psychology
39 (5): 752–66. doi:10.1037/0022-3514.39.5.752.
Chaiken, S., and D. Maheswaran. 1994.“Heuristic
Processing Can Bias Systematic Processing: Effects of
Source Credibility, Argument Ambiguity, and Task
Importance on Attitude Judgment.”Journal of Personality
and Social Psychology 66 (3): 460–473. doi:10.1037/0022-
3514.66.3.460.
Chen, Z., and N. H. Lurie. 2013.“Temporal contiguity and
Negativity Bias in the Impact of Online Word of Mouth.”
Journal of Marketing Research 50 (4): 463–476. doi:10.
1509/jmr.12.0063.
Childers, C. C., L. L. Lemon, and M. G. Hoy. 2019.
“#Sponsored #Ad: Agency Perspective on Influencer
Marketing Campaigns.”Journal of Current Issues &
Research in Advertising 40 (3): 258–274. doi:10.1080/
10641734.2018.1521113.
De Veirman, M., V. Cauberghe, and L. Hudders. 2017.
“Marketing Through Instagram Influencers: The Impact
of Number of Followers and Product Divergence on
Brand Attitude.”International Journal of Advertising 36
(5): 798–828. doi:10.1080/02650487.2017.1348035.
Ellet, John. 2018.“New Research Shows Growing Impact
of Online Research on In-Store Purchases.”Forbes,
February 8. https://www.forbes.com/sites/johnellett/2018/02/
08/new-research-shows-growing-impact-of-online-research-
on-in-store-purchases/#ce4c4f716a0b
Evans, S. K., K. E. Pearce, J. Vitak, and J. W. Treem. 2017.
“Explicating Affordances: A Conceptual Framework for
Understanding Affordances in Communication
Research.”Journal of Computer-Mediated Communication
22 (1): 35–52. doi:10.1111/jcc4.12180.
Festinger, L. 1962.“Cognitive Dissonance.”Scientific
American 207 (4): 93–102. https://www.jstor.org/stable/
24936719. doi:10.1038/scientificamerican1062-93.
Filieri, R., E. Raguseo, and C. Vitari. 2021.“Extremely
Negative Ratings and Online Consumer Review
Helpfulness: The Moderating Role of Product Quality
Signals.”Journal of Travel Research 60 (4): 699–717. doi:
10.1177/0047287520916785.
Fischer, P., E. Jonas, D. Frey, and S. Schulz-Hardt. 2005.
“Selective Exposure to Information: The Impact of
Information Limits.”European Journal of Social
Psychology 35 (4): 469–92. doi:10.1002/ejsp.264.
Fiske, S. T., and S. E. Taylor. 1991.McGraw-Hill Series in
Social Psychology. Social Cognition. 2nd ed. New York:
Mcgraw-Hill Book Company.
Flynn, L. R., and R. E. Goldsmith. 1999.“A short, Reliable
Measure of Subjective Knowledge.”Journal of Business
Research 46 (1): 57–66. doi:10.1016/S0148-2963(98)00057-5.
Fogg, B. J., G. Cuellar, and D. Danielson. 2002.“Motivating,
Influencing, and Persuading Users.”In The human-
Computer Interaction Handbook: Fundamentals, Evolving
Technologies and Emerging Applications, edited by J. A.
Jacko and A. Sears, 133–147. Hillsdale, NJ: Erlbaum.
Fu, W. W., and C. C. Sim. 2011.“Aggregate Bandwagon
Effect on Online Videos’Viewership: Value Uncertainty,
Popularity Cues, and Heuristics.”Journal of the American
Society for Information Science and Technology 62 (12):
2382–2395. doi:10.1002/asi.21641.
Gass, R. H., and J. S. Seiter. 2011.“Credibility.”In
Persuasion: Social Influence and Compliance Gaining,
JOURNAL OF INTERACTIVE ADVERTISING 17

edited by K. Bowers, J. Zalesky, and M. Lentz. 4th ed.,
72–90. New York, NY: Pearson
Gigerenzer, Gerd, and Wolfgang Gaissmaier. 2011.
“Heuristic Decision Making.”Annual Review of
Psychology 62:451–482.