See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360603543

Influencer Marketing Effectiveness

Article in Journal of Marketing · May 2022

DOI: 10.1177/00222429221102889

CITATIONS READS

105 40,052

5 authors, including:

Flora F. Gu Yiwei Li
The Hong Kong Polytechnic University Lingnan University
49 PUBLICATIONS 1,976 CITATIONS 25 PUBLICATIONS 230 CITATIONS

SEE PROFILE SEE PROFILE

Jonathan Z. Zhang Robert Palmatier

Colorado State University University of Washington Seattle
34 PUBLICATIONS 940 CITATIONS 146 PUBLICATIONS 17,125 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jonathan Z. Zhang on 03 September 2022.

The user has requested enhancement of the downloaded file.

Article

Journal of Marketing
1-23
Influencer Marketing Effectiveness © American Marketing Association 2022
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/00222429221102889
journals.sagepub.com/home/jmx

Fine F. Leung , Flora F. Gu , Yiwei Li, Jonathan Z. Zhang,

and Robert W. Palmatier

Abstract
Influencer marketing initiatives require firms to select and incentivize online influencers to engage their followers on social media
in an attempt to promote the firms’ offerings. However, limited research considers the costs of influencer marketing when eval-
uating these campaigns’ effectiveness, particularly from an engagement elasticity perspective. Moreover, it is unclear whether and
how marketers could enhance influencer marketing effectiveness by strategically selecting influencers, targeting their followers, or
managing content. This study draws on a communication model to examine how factors related to the sender of a message (influ-
encer), the receiver of the message (influencer’s followers), and the message itself (influencer’s posts) determine influencer mar-
keting effectiveness. The findings show that influencer originality, follower size, and sponsor salience enhance effectiveness, and
posts that announce new product launches diminish it. Several tensions arise when firms select influencers and manage content:
Influencer activity, follower–brand fit, and post positivity all exert inverted U-shaped moderating effects on influencer marketing
effectiveness, suggesting that firms that adopt a balanced approach along these dimensions can achieve greater effectiveness.
These novel insights offer important implications for marketers designing influencer marketing campaigns.

Keywords
influencer marketing effectiveness, online influencers, consumer engagement, social media, marketing strategy
Online supplement: https://doi.org/10.1177/00222429221102889

Consumers’ growing skepticism toward traditional marketing Over 75% of marketers intend to dedicate resources to influ-
has made it increasingly difficult for firms to attract and influ- encer marketing, and related spending is expected to reach
ence consumers. Many marketers turn to online influencers to $16.4 billion by the end of 2022 (Influencer Marketing Hub
promote their brands and products on social media (e.g., 2022). However, industry reports predict that “there are disap-
Instagram, Facebook, Weibo), propelling the growth of “influ- pointed marketers out there spending budgets but not really
encer marketing,” a communication strategy in which a firm knowing what benefit the campaign has brought them”
selects and incentivizes online influencers to engage their fol- (Brennan 2019). Influencer marketing requires a lot of resources
lowers on social media in an attempt to promote the firm’s but also is difficult to implement and assess, making it critical to
offering (Leung, Gu, and Palmatier 2022). The firms select identify the decision criteria that firms can use to enhance the
and pay (e.g., pay-per-post) online influencers—individuals, effectiveness of their influencer marketing efforts. According
groups of individuals, or even virtual avatars who have built to emerging research, certain features of influencer marketing,
networks of followers on social media (De Veirman, such as source and post characteristics, affect outcomes such
Cauberghe, and Hudders 2017). Although some influencers
build such a large following that they attain celebrity status,
they differ from celebrities in the source of their fame. Fine F. Leung is Assistant Professor of Marketing, Faculty of Business, Hong
Kong Polytechnic University, Hong Kong (email: fi[email protected]).
Whereas celebrities have succeeded in some credentialed, Flora F. Gu is Associate Professor of Marketing, Faculty of Business, Hong
institutional setting (e.g., acting, music, sports), influencers Kong Polytechnic University, Hong Kong (email: fl[email protected]).
are not certified by formal institutions (McQuarrie, Miller, Yiwei Li is Assistant Professor, Department of Marketing and International
and Phillips 2013). They accumulate followers by sharing Business, Lingnan University, Hong Kong (email: [email protected]).
content, and they weave brand endorsements into their per- Jonathan Z. Zhang is Associate Professor of Marketing, College of Business,
Colorado State University, USA (email: [email protected]).
sonal stories and posts, resulting in content that appears Robert W. Palmatier is Professor of Marketing, John C. Narver Chair in
authentic and provides consumption value (Lou and Yuan Business Administration, University of Washington, USA (email: palmatrw@
2019). uw.edu).
2 Journal of Marketing 0(0)

as consumer engagement (Hughes, Swaminathan, and Brooks includes each post’s sender (i.e., influencer), receiver (i.e., fol-
2019; Valsesia, Proserpio, and Nunes 2020), brand and influ- lowers), and message characteristics. This data set is unique
encer attitudes (De Veirman, Cauberghe, and Hudders 2017; in several ways compared with the data that support prior
De Veirman and Hudders 2020), purchase intentions (Lee and studies (e.g., Hughes, Swaminathan, and Brooks 2019). It fea-
Eastin 2020; Lou and Yuan 2019), and product sales tures a more diverse group of online influencers, spanning a
(Bharadwaj et al. 2022). However, few studies explicitly broader range of campaigns, brands, and categories (i.e.,
assess influencer marketing effectiveness in terms of engage- 5,835 influencer marketing posts related to 1,256 campaigns
ment elasticity, defined as the percentage change in consumer written by 2,412 influencers, sponsored by 861 brands in 29 cat-
engagement due to a 1% increase in influencer marketing egories). With access to influencer cost data for each post, as
spend.1 This gap might reflect the lack of access to influencer well as 24-hour lagged engagement data (e.g., number of
cost data, but ignoring such costs hinders any accurate evalua- reposts), we can estimate influencer marketing effectiveness in
tion of the effectiveness of marketing spending. As influencer terms of engagement elasticity to establish the incremental con-
marketing becomes increasingly competitive, firms’ ability to tribution that influencer marketing spend makes for fostering
allocate their budgets optimally, by selecting individual influ- engagement.
encers and managing individual posts in ways that maximize This research contributes to extant literature in several ways.
engagement elasticity, can establish their competitive advan- First, as mentioned, recent studies note various consumer and
tages. We thus gather influencer cost and engagement data firm outcomes of influencer marketing (e.g., Bharadwaj et al.
and undertake a systematic assessment of influencer marketing 2022; Hughes, Swaminathan, and Brooks 2019) (Table 1), but
effectiveness across varied conditions. most of them do not account for the costs of generating those
Our proposed conceptual framework reflects communication outcomes. As Batra and Keller (2016, p. 136) argue, “marketers
models (Lasswell 1948; Shannon and Weaver 1949) and their must evaluate marketing communications … against their cost
component characteristics, related to (1) the sender of a to arrive at the most effective and most efficient communica-
message (influencer in our research context), (2) the receiver tions program.” Therefore, this study empirically examines
of the message (influencer’s followers), and (3) the message the effectiveness of influencer marketing spend (i.e.,
itself (influencer’s marketing post). Specifically, we investigate pay-per-post) for generating consumer engagement, as mea-
whether engagement elasticity is moderated by selecting influ- sured by engagement elasticity. Our results reveal that
encers who post more or fewer posts (influencer activity), increasing the influencer marketing budget can increase con-
provide original content (originality), or have more or fewer fol- sumer engagement: Ceteris paribus, a 1% increase in influ-
lowers (follower size); targeting follower networks with differ- encer marketing spend increases engagement by .457%. By
ent levels of follower–brand fit; and posting content with assessing and comparing influencers’ engagement elasticities
distinct degrees of post positivity and sponsor salience, or and base engagement levels, we also establish how firms can
content that relates to new product launches (see Figure 1). allocate their budgets optimally. On average, the firms in our
From this framework, we derive two research questions: data set could increase consumer engagement by 16.6% if
they allocated their budgets proportional to these elasticities
and base engagement levels, rather than their current
1. Does consumer engagement increase in response to
allocations.
influencer marketing spend?
Second, we apply a communication model (Lasswell 1948;
2. Can marketers enhance engagement elasticity by strate-
Shannon and Weaver 1949) as an alternative to the theories
gically selecting influencers and their followers, as well
adopted in prior research (Table 1), which offers a more com-
as managing their content? That is, how do influencer-,
prehensive assessment of how factors related to the sender
follower-, and post-related factors interact with influ-
(influencer), receiver (followers), and message (influencer mar-
encer marketing spend in affecting consumer engage-
keting post) lead to varied influencer marketing effectiveness.
ment of the sponsored post?
These categories of factors are central to campaign designs,
which generally require selecting effective influencers and fol-
To address these questions, we obtained data from a large influ- lower groups to target, as well as defining effective posts.
encer marketing platform (Data Provider) that enables firms to Selecting influencers who transmit more original posts than
select and pay online influencers to share sponsored posts posts created by others and with larger networks of followers,
about their brands and products on various social networks. along with incorporating more clickable mentions and links in
We gather rich data about sponsored posts, which appeared the sponsored posts, enhances effectiveness. Sponsored posts
on a prominent social network and were transacted through that announce new product launches diminish effectiveness
the Data Provider in October 2018; the gathered information because of the potential risks and advertising clutter involved
with new products. This overarching communication model
also sheds new light on follower–brand fit, a relatively less
1
We use “influencer marketing effectiveness” and “engagement elasticity”
interchangeably to refer to the main effect of influencer marketing spend on con- studied receiver factor, and reveals the promising potential of
sumer engagement. In the conceptual framework, we detail factors that moderate leveraging big data to make effective targeting decisions
this main effect (Figure 1). (Nelson and Webster 2016).
Leung et al. 3

Figure 1. Conceptual framework of influencer marketing effectiveness.

Third, influencer marketing agreements empower influencers spend in a particular medium) increases by 1% (Danaher and
to transmit brand-related information to target consumers, Van Heerde 2018). Elasticity provides a suitable measure for
which differs from traditional brand- or user-generated gauging advertising effectiveness because it is dimensionless
content. This research establishes evidence of inverted and can be computed for any outcome variable (Venkatraman
U-shaped moderating effects of influencer activity, follower– et al. 2015). Prior studies that examine the advertising elastici-
brand fit, and post positivity. In turn, we suggest that firms ties of various traditional (e.g., print, radio) and online (e.g.,
should select influencers who display medium levels of paid search, online display) advertising media also identify con-
posting activity rather than those who post too frequently. textual factors that alter these elasticities (Becker, Wiegand, and
This insight also helps reconcile some mixed findings in previ- Reinartz 2019; Datta, Ailawadi, and Van Heerde 2017; Van
ous studies (Stephen et al. 2017; Suh et al. 2010). A prevailing Heerde et al. 2013).
view in celebrity endorsement literature suggests that brand fit is However, few prior studies account for the costs associated
a strong indicator of effective communication (Bergkvist and with online influencers to establish their effectiveness or elastic-
Zhou 2016), but we determine that followers with a high ity. Therefore, we sought access to unique spending data for
degree of shared interests with the brand may not be the best influencer marketing posts. Then, to reflect a primary objective
group to target with influencer marketing. Finally, adding of influencer marketing—namely, to encourage consumers’
nuance to previous literature that suggests positive content is engagement with sponsored content on social media (Hughes,
more viral (Berger and Milkman 2012), our analysis indicates Swaminathan, and Brooks 2019)—we include engagement as
that a blend of positive and negative content can increase an outcome variable. Broadly defined to encompass cognitive,
engagement elasticity by 5.6% (22.4%), compared with emotional, and behavioral activities (Hollebeek, Glynn, and
content that is one (two) standard deviation(s) higher in positiv- Brodie 2014), engagement on social media can be operational-
ity. These nonlinear effects help clarify some unique tensions ized as a set of measurable consumer behaviors in response to
that arise in influencer marketing campaigns: Followers know online content, such as liking, commenting, or reposting
influencers are paid, but influencers still need to appear authen- content (Malhotra, Malhotra, and See 2013). These forms of
tic, display leadership, and provide communication value to engagement create ripple effects, influence other potential cus-
those followers (Leung, Gu, and Palmatier 2022). tomers, and contribute to firm performance (Pansari and
Kumar 2017). We prioritize the number of reposts an influencer
marketing post generates because reposting is a deeper form of
Conceptual Background and Hypothesis engagement than just liking; it implies consumers self-select to
Development propagate the content to their own networks (Malhotra,
Influencer Marketing Effectiveness Malhotra, and See 2013).

Considerable marketing literature deals with the effectiveness

with which advertising can produce relevant firm outcomes
such as sales, market share, and firm value (Dinner, Van
Models of Communication
Heerde, and Neslin 2014; Sethuraman, Tellis, and Briesch Similar to other communication strategies, firms use influencer
2011; Sridhar et al. 2016). A conventional measure of advertis- marketing to communicate and deliver value to consumers, in
ing effectiveness is elasticity, defined as the percentage increase the pursuit of favorable firm outcomes (Leung, Gu, and
in an outcome variable when an input variable (e.g., advertising Palmatier 2022). Yet instead of communicating directly with
Table 1. Review of Influencer Marketing Literature.

Research Accounts for Influencer Follower

Authors Context Costs Factors Factors Post Factors Other Factors Outcomes Theory Key Findings
A MO
Bharadwaj et al. Videos of live No Not studied Not studied Face presence ; Product price ; Sales Emotions as social Emotional displays in
(2022) sales emotional promotionMO information livestream retailing have
presentations displaysA model negative U-shaped effects on
sales over time.
Breves et al. Experiment and No Influencer–brand Not studied Not studied Parasocial Perceived Social adaptation; Influencer–brand fit positively
(2019) survey about fitA relationshipsMO credibility; brand attribution affects influencer credibility
Instagram evaluation; theory and ad outcomes, especially
influencers behavioral for followers with weak
intentions parasocial relationships.
De Veirman, Experiments No Follower sizeA; Not studied Not studied Product Influencer Heuristic Influencers with more
Cauberghe, and about Instagram perceived divergenceMO likeability processing; naive followers are perceived as
Hudders (2017) influencers popularityME; theories more popular and likable;
opinion the effects depend on
leadershipME; number of followers and
number of product divergence.
followeesMO
De Veirman and Experiment about No Source Not studied Sponsorship SkepticismME Brand attitude Persuasion Sponsorship disclosure
Hudders (2020) Instagram credibilityME disclosureA; ad knowledge negatively affects brand
influencers recognitionME; model; attitude through enhanced
message reactance theory ad recognition and
sidednessMO skepticism, which lowers
influencer credibility,
especially when using a
one-sided message.
Gong et al. (2017) Field experiment Yesa Follower sizeMO; Not studied Tweet vs. tweet Number of Show viewing — Influencer retweets increase
with a media daily number of and retweetA company viewing if the show tweet is
company and tweetsMO followersME informative; they are more
influencers on effective than company
Weibo tweets in bringing new
followers, which indirectly
increases viewing.
Hughes, Sponsored Yesb Influencer Not studied Hedonic valueA Campaign Engagement Elaboration Effects of sponsored blogging
Swaminathan, influencer posts expertiseA incentiveA; likelihood model on engagement depend on
and Brooks from campaign influencer characteristics
(2019) Motherhood intentMO; and post content, which are
platform typeMO further moderated by
platform type and campaign
intent.
Lanz et al. (2019) Music creators No Status differenceA Not studied Not studied Not studied Follow-backs Seeding strategies; Responsiveness of seeding
from a leading social identity targets declines with status
music platform theory difference; unknown music
creators benefit from
targeting low-status users
rather than influencers.
Lee and De Online survey with No Influencer Not studied Not studied Perceived Strength of brand Stereotype change Influencer typicality shapes
Fortuny (2022) social media typicalityA homogeneity of reference group perceived homogeneity of
users brand associations brand consumers, which
consumersME affects the strength and
tightness of brand
associations.

(continued)
Table 1. (continued)
Research Accounts for Influencer Follower
Authors Context Costs Factors Factors Post Factors Other Factors Outcomes Theory Key Findings

Lee and Eastin Experiment about No Perceived Not studied Not studied Consumer Attitude toward Brand personality; Influencer sincerity positively
(2020) Instagram sincerityA envyMO; product influencer; brand schema theory affects consumer attitudes;
MO
influencers type attitude; the effects are contingent on
purchase consumer envy and product
intention types.
Lou and Yuan Online survey with No TrustworthinessA; Not studied Informative valueA — Perceived trust; Source credibility; Informative value of influencer
(2019) social media attractivenessA; brand advertising content, influencer
users similarity to awareness; content value trustworthiness,
followersA purchase attractiveness, and similarity
intentions to the followers positively
affect follower trust, which
influence brand awareness
and purchase intentions.
Pei and Mayzlin Bayesian Yesc Not studied Prior beliefMO AffiliationA Cost of Value of Persuasion theory Affiliation decision depends
(2021) persuasion information information on the cost of information
model acquisitionMO; acquisition, consumers’
disclosure prior belief, and disclosure
regimeMO regime.
Valsesia, Twitter posts; lab No Follower sizeMO; Not studied Not studied — Engagement Heuristic Following fewer others
Proserpio, and experiments number of processing conveys greater autonomy,
Nunes (2020) about followeesA; which positively affects
influencers perceived perceived influence and
autonomyME; engagement.
perceived
influenceME
Yuan and Lou Online survey with No Perceived Not studied Not studied Parasocial Product interest Source credibility; Influencer credibility and
(2020) social media credibilityA; relationshipME communication fairness positively affect the
users perceived justice strength of parasocial
fairnessA relationships with
influencers, which increases
followers’ interests in
influencer-promoted
products.
This article Sponsored Yes (influencer Influencer Follower–brand Post positivityMO; — Engagement Communication Factors related to influencers,
influencer posts marketing activityMO; fitMO sponsor model their followers, and their
on Weibo spendA) influencer salienceMO; new sponsored posts have
originalityMO; product critical roles in determining
follower sizeMO launchMO influencer marketing
effectiveness in terms of
engagement elasticity.

Notes: A = antecedent; ME = mediator; MO = moderator.

a
Each influencer was paid 1,000 CNY for their participation in a field experiment. The authors computed the returns on tweeting across experimental conditions.
b
The cost for each blogger campaign was considered in a post hoc analysis. The authors calculated return on engagement by dividing the total revenue generated by the total cost for each blogger campaign; they find
significant positive effects of blogger expertise and campaign incentives on return on engagement.
c
In the model setup, the firm may manipulate the review process by offering the influencer payment in exchange for affiliation.
6 Journal of Marketing 0(0)

consumers, influencer marketing requires firms to empower (Barker 2018). Followers may then grow uninterested in the
influencers to create and transmit brand-related information influencer’s posts, selectively filter them, or even feel annoyed
through social media. This transmission is what determines by them, making the influencer a less effective message sender.
the process of value communication specific to influencer mar- These predictions suggest that, after an optimal point, an influ-
keting (Balducci and Marinova 2018), which also may align encer with greater posting activity lowers the effectiveness of
with traditional communication models (Lasswell 1948; influencer marketing spend because followers become less
Shannon and Weaver 1949) that depict processes by which mes- responsive to that influencer’s individual posts. Whereas select-
sages flow from senders to receivers. The sender is the source of ing an influencer who engages in minimal activity may not
the message. The message, consisting of words, sounds, or produce perceptions of credibility as a message sender, selecting
behaviors, gets transmitted through a channel to a receiver, one with excessive activity may backfire as a result of possible
who is the audience. Most models thus cite three common ele- information overload. Therefore, we predict that influencer activ-
ments that shape effective communication: (1) the sender of a ity first strengthens the positive effect of influencer marketing
message, (2) the receiver of the message, and (3) the message spend on consumer engagement; yet, after reaching an optimal
itself (Swani, Brown, and Milne 2014; Walker et al. 2017). point, it starts to weaken the effect. That is, a moderate level of
The sender’s characteristics indicate how believable or influen- influencer activity is optimal for producing the highest engage-
tial they are as a message source (Self 2009; Wilson and Sherrell ment elasticity.
1993); receivers’ characteristics determine how involved they
will be with the message (Boerman, Van Reijmersdal, and H1: Influencer activity has an inverted U-shaped effect on
Neijens 2015; Eisend and Tarrahi 2016); and the message’s influencer marketing effectiveness.
characteristics denote its content value (Ducoffe 1996; Lou
and Yuan 2019). Thus, all these factors can enhance or In managing their social media accounts, influencers post
detract from the message’s potential to elicit responses from content written or produced by themselves or created by
receivers. Accordingly, we predict that characteristics pertain- others. The tendency to post one type of content over another
ing to the influencer (sender), the influencer’s followers reflects influencer originality, defined as the degree to which
(receiver), and the influencer’s marketing post (message) func- online influencers create original content on social media and
tion as moderators that lead to the varying effectiveness (i.e., thereby achieve differentiation in followers’ minds (Casaló,
engagement elasticity) of influencer marketing. Flavián, and Ibáñez-Sánchez 2020). We expect influencer orig-
inality to enhance influencer marketing effectiveness for several
reasons. First, original influencers produce content in their own
Sender (Influencer) Characteristics words and style, which makes them stand out from the crowd
Influencers are fundamentally content generators and dissemi- of influencers in the market. Consumers like to talk or read
nators on social media (Yuan and Lou 2020). Unlike firm- about topics they find interesting or surprising (Moldovan,
generated communications, for which firms send brand mes- Goldenberg, and Chattopadhyay 2011), so influencers who
sages, influencer marketing empowers influencers to take on offer greater originality should attract more attention than
sender roles. Firms tend to select influencers on the basis of those with low originality. Second, influencers who share orig-
their posting behaviors, such as how frequently or what kind inal content and ideas are likely to be perceived as knowledge-
of content they tend to post, and their follower size. We accord- able, credible senders with whom followers may prefer to
ingly examine how influencer activity, originality, and follower interact (Ki and Kim 2019). Third, originality is a defining prop-
size, as sender traits, may moderate engagement elasticity. erty of authenticity (Nunes, Ordanini, and Giambastiani 2021)
Influencer activity refers to the frequency with which an and a key trait of successful influencers (Casaló, Flavián, and
influencer transmits content (e.g., messages, photos, videos) Ibáñez-Sánchez 2020). Original influencers use personalized
on social media (Stephen et al. 2017). When influencers post methods to show how the touted product fits into their everyday
frequently, followers infer that the information sent by influ- lives, which consumers perceive as trustworthy (Leung, Gu, and
encers is fresh and up-to-date (Stephen et al. 2017). Palmatier 2022). Overall, original influencers may be more
Moreover, because influencers frequently post on social likely to be noticed and trusted by consumers, which makes
media, followers come to believe that they know the influencers them more effective message senders. We expect influencer
intimately (Escalas and Bettman 2017), which enhances their originality to positively interact with influencer marketing
trust in the influencer. An increase in spend on an influencer spend; that is, increasing the spend on original influencers
who posts more frequently should generate more engagement, should generate more engagement than increasing the spend
because the followers perceive the influencer as an updated, on those who are less original.
trustworthy message sender and thus are more likely to
respond (Wilson and Sherrell 1993). H2: Influencer originality enhances influencer marketing
However, if an influencer posts extremely frequently, the effectiveness.
large volume of posts may distract followers and dilute their
attention to any particular post (Gong et al. 2017). Posting too Follower size refers to the number of followers an online influ-
frequently also can clutter followers’ feeds and create fatigue encer has on a social media platform (De Veirman, Cauberghe,
Leung et al. 7

and Hudders 2017). This characteristic provides an important high if a cosmetic brand were to work with an influencer
criterion for selecting influencers because it is easily observable whose followers are interested in beauty, medium if those fol-
on most platforms. Although most social network studies identify lowers are interested in beauty and food, and low if they are
social hubs (i.e., well-connected people with many connections to interested only in food. In influencer marketing contexts, fol-
others) as favorable seeding targets because their connectivity lowers’ digital consumer profiles provide a means for brands
enables wider propagation and greater market size (Goldenberg to engage in audience targeting (Neumann, Tucker, and
et al. 2009; Hinz et al. 2011; Libai, Muller, and Peres 2013), it Whitfield 2019). Firms can use consumers’ viewing and click-
is unclear whether having more followers creates a stronger ing data to understand their interests almost instantaneously as
response to an increase in influencer spend on engagement. they surf the internet. As such, follower–brand fit is widely
Some practitioners suggest that working with influencers with employed by influencer marketers (Hobbs 2019) and appears
smaller followings is worthwhile because they may be perceived key to unlocking the potential of big data for improving market-
as more relatable and authentic (Hosie 2019); however, we posit ing communications (Varnali 2021).
that soliciting influencers with large follower size is effective for When a brand’s domain aligns with the interests of the
fostering engagement elasticity for two reasons. influencer’s followers, a sponsored post about the brand is
First, a large follower network grants an influencer access to more personally relevant to those followers (Geng et al. 2021),
a sizable pool of potential consumers and stronger potential for which should motivate them to process the information (Petty
eliciting engagement responses. This is analogous to market and Cacioppo 1986). That is, these followers are relevant receiv-
entry contexts in which potential entrants often speculate that ers to target because they are likely to become involved with the
a large market size may offer them a better chance of success post, evaluate it carefully, and determine whether and how to
(Min, Kim, and Zhan 2017). Because having many followers respond (Boerman, Van Reijmersdal, and Neijens 2015).
grants an influencer greater potential reach, devoting more of Because follower–brand fit evokes followers’ interests and
the budget to collaborating with this influencer should generate draws their attention, an increase in spend on an influencer
more engagement and increase engagement elasticity. Second, whose followers display high (vs. low) brand fit should elicit
more followers also evoke credibility. Follower size serves as more engagement, leading to greater engagement elasticity.
a signifier of the influencer’s popularity (De Veirman, However, if follower–brand fit is already high, increasing it
Cauberghe, and Hudders 2017), status, and reputation further may backfire for several reasons. First, the influencer’s
(Labrecque et al. 2013). When an influencer has a large follow- followers already consume substantial social media content
ing, consumers likely believe the influencer is a valid and reli- related to their interests. If a post closely matches an interest,
able message sender, ascribing greater opinion leadership and it competes for followers’ attention with a clutter of other similar
source credibility to them (Goldenberg et al. 2009). As a content (Nan and Faber 2004), which reduces the chances that it
result, spending more to work with this influencer should be will be noticed or processed. Second, consumers become satiated
more effective because consumers likely pay more attention with specific topics when the related content exceeds a certain
to and more actively repost the content posted by a credible level, after which they seek variety (McAlister 1982); if an influ-
sender (Self 2009; Wilson and Sherrell 1993). Because of encer instead posts novel marketing information with some
both reach and credibility effects, we expect that influencers unmatched elements, it may break the monotony and draw follow-
with more followers are more effective message senders. ers’ attention (Ordenes et al. 2018). Third, when consumers
Therefore, follower size should positively interact with influ- receive appeals closely matched to their interests, they may
encer marketing spend to generate consumer engagement, grow suspicious that the content is commercially motivated or
prompting greater engagement elasticity for firms. manipulative. This growing suspicion can elicit consumer reac-
tance (Campbell and Kirmani 2000), which likely hampers the
H3: Influencer follower size enhances influencer marketing effectiveness of the brand’s influencer marketing spend.
effectiveness. Therefore, we expect that followers with an intermediate level
of follower–brand fit represent the optimal receiver group to
target because doing so leverages their shared interests but
avoids information clutter, satiation, or reactance. We posit
Receiver (Follower) Characteristics that follower–brand fit strengthens the positive effect of influ-
Firms also select influencers on the basis of their followers’ encer marketing spend on consumer engagement up to a
characteristics, according to whether those followers, as receiv- certain point, beyond which it starts to weaken the effect.
ers of a sponsored post, are likely to find the content valuable
and exhibit greater potential for engagement (Swani, Brown, H4: Follower–brand fit has an inverted U-shaped effect on
and Milne 2014). Marketers often consider the composition of influencer marketing effectiveness.
an influencer’s follower network, so we focus on follower–
brand fit to predict how it may lead to variations in influencer
marketing effectiveness. Follower–brand fit is the degree to Message (Post) Characteristics
which the interests of an influencer’s followers match the Firms grant influencers a great deal of freedom to generate
domain of the sponsor brand. For example, this fit would be content, yet firms still manage the content by providing briefs
8 Journal of Marketing 0(0)

that outline the campaign’s objectives or key messages (Leung, sponsor brand is salient increases a post’s commercialism
Gu, and Palmatier 2022). Thus, we investigate how firms may (Tellis et al. 2019). The more an influencer’s post links to the
leverage post positivity, sponsor salience, and new product sponsor brand, the more it looks like a traditional advertisement
launch information to vary influencer marketing effectiveness. and less like an authentic sharing from the influencer (Stubb
Post positivity is the degree to which an influencer marketing 2018). Similar to intense branding activity in TV commercials,
post is positive (Berger and Milkman 2012). We anticipate an salient links to the sponsor brand make a post seem too “hard
inverted U-shaped moderating effect on engagement elasticity. sell,” which may annoy consumers (Bruce, Becker, and
Any sponsored post is likely positive by nature (Haenlein et al. Reinartz 2020; Teixeira, Wedel, and Pieters 2010). Increasing
2020) because positivity indicates the influencer’s endorsement spend on such content should be less effective, suggesting
and persuasive attempt (Akpinar and Berger 2017). When an that sponsor salience will negatively interact with influencer
influencer endorses the brand, consumers anticipate that the offer- marketing spend in eliciting consumer engagement. Given the
ing has some valuable features that also may be worth sharing competing arguments on the moderating role of sponsor sali-
(Barasch and Berger 2014). Influencer marketing spend devoted ence, we propose the following competing hypotheses:
to encouraging more positive posts thus should be beneficial,
but we also note that the inherent features of influencer marketing
may diminish these benefits if positivity builds beyond a certain H6: Sponsor salience (a) enhances influencer marketing effec-
point. Unlike organic recommendations, consumers know that tiveness or (b) reduces influencer marketing effectiveness.
influencers are paid to advocate for the brand. When influencers
share highly positive comments about the brands they endorse
(i.e., post positivity is high), followers may question the extent
to which their post is authentic (McQuarrie, Miller, and Phillips Finally, brands often solicit influencers to post about a new
2013) and instead perceive manipulative intent. Manipulative product launch. These posts are typically aimed at increasing con-
tactics can reduce perceived content value (Ducoffe 1996), acti- sumer awareness about the new products and encouraging the
vate consumers’ persuasion knowledge, and elicit reactance spread of information on social media (Hughes, Swaminathan,
(Campbell and Kirmani 2000). A moderately positive post that and Brooks 2019). We expect that consumers should pay more
includes some negative comments should offer greater content heed to influencer posts introducing new products because they
value than posts with low or high levels of positivity (Uribe, contain more novel and interesting information than those mar-
Buzeta, and Velásquez 2016), because it is positive enough to keting existing products (Berger 2014). Prior literature suggests
generate positive impressions but still allow for perceptions of dis- that new product advertisements are generally shared more
cernment. We thus expect an increase in influencer marketing because they make the sharers appear knowledgeable about the
spend on moderately positive posts to produce the most engage- marketplace (Tellis et al. 2019). Accordingly, consumers in our
ment, resulting in the highest level of engagement elasticity. study may more actively engage with or share new product
launch posts to express their uniqueness or fulfill their socializing
H5: Post positivity has an inverted U-shaped effect on influ- and helping motives (Berger and Schwartz 2011; Tellis et al.
encer marketing effectiveness. 2019), enhancing influencer marketing effectiveness.
However, new product launches also involve significant risks
Sponsor salience, or the extent to which the sponsor brand is prom- (Tellis et al. 2019); nearly 95% of them fail (Hyder 2019).
inent in a post (Teixeira, Wedel, and Pieters 2010; Tellis et al. Because consumers face uncertainty related to the performance,
2019), affects how consumers make inferences about the social desirability, and appropriateness of new products
content. We operationalize it as the total count of sponsor (Castaño et al. 2008), engaging with or reposting new product
brand-related mentions (or @s) and URL links, and we provide announcements may appear risky and contrary to their desire to
competing predictions on its moderating effect on engagement avoid social disapproval (Berger 2014). In support of this reason-
elasticity. According to conversational norms, contributions to a ing, a study of Facebook advertisements (Gavilanes, Flatten, and
conversation should be informative (Grice 1975), and that norm Brettel 2018) reveals that people hesitate to share content about
should apply to influencer marketing too. Prior studies note the new products they have not yet experienced. An increase in
importance of online content’s practical utility for encouraging spend on influencer posts that announce new product launches,
engagement (Berger and Milkman 2012; Rooderkerk and rather than those that market existing products, should generate
Pauwels 2016). When the sponsor brand is prominent in a post, less engagement because the heightened risks linked to new prod-
it offers important information to consumers that helps them com- ucts likely reduce consumers’ responsiveness to posts that feature
prehend and learn from the content (Teixeira, Wedel, and Pieters such information. Considering the competing arguments, we
2010). Influencer posts with higher sponsor salience are therefore offer the final set of competing hypotheses regarding the moder-
perceived to be more informative and possess greater content value ating role of new product launch posts:
(Lou and Yuan 2019), and increasing spend to encourage such
content should elicit greater consumer engagement.
However, there are also reasons to believe that sponsor sali- H7: New product launch posts (a) enhance influencer marketing
ence may reduce engagement elasticity. The degree to which the effectiveness or (b) reduce influencer marketing effectiveness.
Leung et al. 9

Data and Measures an influencer published on Weibo in the 90 days prior to the cam-
paign (Gong et al. 2017; Stephen et al. 2017). Higher activity
Data implies that the influencer is a more active sender of information
We test our framework using data obtained from a publicly on social media. Influencer originality is the ratio of the number of
listed influencer marketing platform in China (Data Provider). original posts (i.e., both sponsored and nonsponsored posts an
It helps brands select and pay online influencers to share posts influencer wrote or produced) to the total number of posts an influ-
about their products on various social networks. Annually, it encer published in the 90 days prior to the campaign. Greater orig-
facilitates more than 120,000 influencer marketing post transac- inality implies that the influencer creates more original content,
tions. Data Provider shared data about its post transactions on rather than transmitting content created by others. Follower size
Weibo (a popular microblogging site in China, similar to is the total number of followers (in millions) an influencer has
Twitter) with us. The data pertain to 5,835 influencer marketing on Weibo, prior to the campaign (Kupfer et al. 2018). It reflects
posts written by 2,412 online influencers related to 1,256 cam- the influencer’s potential reach and perceived credibility.
paigns for 861 brands in October 2018. The sponsor brands
span 29 categories; the 6 largest are beauty products, e-com- Receiver (follower) characteristic. To measure follower–brand fit,
merce platforms, food and beverages, electronics, apparel, and we examine the degree to which the interests of an online influ-
personal care products. The unique data set includes each encer’s followers match the associated domains of a post’s
post’s cost and 24-hour lagged engagement data (i.e., number sponsor brand. Weibo identifies users’ interest(s) (e.g.,
of reposts and comments captured 24 hours after the post was travel, beauty, fashion, music) on the basis of their social
shared on Weibo), so we can probe cost-based effectiveness media activities on the site. For each influencer, we obtained
in driving engagement. The substantial variance in size and data on the number of followers identified as interested in
focus of the influencers, their followers, and their posts each of the 42 domains prior to the campaign. Then, two inde-
support our analysis of contingencies. pendent coders coded whether a sponsor brand is associated
with each of the same 42 interest domains (e.g., Lancôme is
Measures associated with beauty). The interrater agreement was
96.8%, and disagreements were resolved by consensus. The
Table 2 details the operationalization of the key variables that measure of follower–brand fit is the percentage of an influ-
map onto our conceptual framework. Table 3 contains the encer’s followers whose interests match the domains of the
descriptive statistics and correlations for the variables. sponsor brand. Greater follower–brand fit indicates that a
brand more accurately targets followers whose behavioral pat-
Engagement. Following prior work (e.g., Valsesia, Proserpio, and terns (on Weibo) reveal interests that align with the brand’s
Nunes 2020), we operationalize engagement, our dependent var- domains.
iable, as the number of reposts an influencer marketing post gen-
erates in its first 24 hours. We use the 24th hour as a cutoff, in
accordance with empirical evidence obtained from Data Message (post) characteristics. We measure post positivity as the
Provider that shows activity largely halts after this point. In con- ratio of the difference between positive and negative words to
versations with marketers, we learned that measuring engagement the total number of emotional words in an influencer marketing
up to this point is consistent with industry standards. To confirm post [(number of positive words − number of negative words)/
the robustness of the findings, we also use the number of com- (number of positive words + number of negative words + 1)].2
ments as of the 24th hour as an alternative dependent variable. We conducted an automated sentiment analysis using a linguis-
tic inquiry and word count dictionary to quantify the number of
Influencer marketing spend. Our focal independent variable is positive and negative words (Pennebaker et al. 2015). Sponsor
influencer marketing spend, or the monetary amount (in salience is measured by the total count of @s to the sponsor
dollars) a brand spends on a sponsored post shared by an influ- brand’s own Weibo account (e.g., @Dior, @KFC) and URLs
encer. The payment is per post, as a one-off cost, such that the linked to the sponsor brand’s website or online shop in an influ-
brand makes the payment to the influencer through Data encer marketing post (Soboleva et al. 2017). Greater sponsor
Provider once they share a post on Weibo. Some brands spend salience suggests a post has more direct links to the sponsor
remarkable amounts (maximum of $92,857) on single posts, brand, which makes the sponsor brand more prominent. We
but the median amount ($293) suggests a reasonable budget. used dictionary-based text analysis to code whether an influ-
Because of substantial variation, we use a log-log model to esti- encer marketing post relates to a new product launch.
mate the effect of this variable on engagement, which yields a Following Humphreys and Wang’s (2018) procedures for
unit-free elasticity measure that accommodates potentially chang- the dictionary creation, dictionary validation, and post-
ing returns to scale and depicts the percentage change in engage- measurement validation steps, we then applied the newly devel-
ment in response to a 1% increase in influencer marketing spend. oped dictionary to a text analysis, for which a post is categorized

Sender (influencer) characteristics. Influencer activity is measured 2

By adding 1 to the denominator (i.e., total number of emotional words), we
as the total number of posts (both sponsored and nonsponsored) ensure the inclusion of posts that contain no emotional words.
10 Journal of Marketing 0(0)

Table 2. Variable Operationalizations.

Variable Definition Operationalization References

Dependent Variable
Engagement Number of reposts generated on an Number of reposts an influencer marketing post Valsesia, Proserpio,
influencer marketing post generated within 24 hours of posting and Nunes (2020)
Independent Variable
Influencer Cost of an influencer marketing post Amount of money (in dollars) spent on an influencer —
marketing marketing post
spend
Sender (Influencer) Characteristics
Influencer Frequency with which an online influencer Total number of posts (both sponsored and Gong et al. (2017);
activity transmits content on social media nonsponsored) an online influencer published on Stephen et al.
Weibo in the preceding 90 days, prior to the (2017)
campaign
Influencer Degree to which an online influencer creates Ratio of number of original posts (both sponsored —
originality original content and non-sponsored) to total number of posts
published in the preceding 90 days, prior to the
campaign
Follower size Number of followers an online influencer has Total number of followers an online influencer has on Hughes,
on a social media platform Weibo (in millions), prior to the campaign Swaminathan, and
Brooks (2019)
Receiver (Follower) Characteristics
Follower–brand Degree to which the interests of an online Percentage of an online influencer’s followers whose —
fit influencer’s followers match with the interests matches with the associated domains of
associated domains of the sponsor brand of the sponsor brand of an influencer marketing post
an influencer marketing post
Message (Post) Characteristics
Post positivity Degree to which the content of an influencer Ratio of the difference between positive and negative Berger and Milkman
marketing post is positive words to the total number of emotional words in an (2012)
influencer marketing post [(number of positive
words − number of negative words)/(number of
positive words + number of negative words + 1)]
Sponsor Degree to which the sponsor brand is Total count of @s (linked to the sponsor brand’s own Soboleva et al. (2017)
salience prominent in an influencer marketing post Weibo account) and URLs (linked to the sponsor
brand’s website or online shop) in an influencer
marketing post
New product Whether an influencer marketing post is Binary variable equal to 1 if an influencer marketing —
launch related to new product launch post is related to new product launch
Control Variables
Promotion Whether an influencer marketing post is Binary variable equal to 1 if an influencer marketing —
related to promotions post is related to promotions
Post length Length of an influencer marketing post Total number of characters in an influencer marketing Berger and Milkman
post (2012)
Post number in Number of influencer marketing posts within Total number of influencer marketing posts within —
campaign the same campaign the same campaign as the focal influencer marketing
post
Service (vs. Whether the sponsor brand of an influencer Binary variable equal to 1 if the sponsor brand of an Lovett, Peres, and
product) marketing post is a service or product brand influencer marketing post is a service (vs. product) Shachar (2013)
brand brand
Premium (vs. Whether the sponsor brand of an influencer Binary variable equal to 1 if the sponsor brand of an Lovett, Peres, and
value) brand marketing post is a premium or value brand influencer marketing post is a premium (vs. value) Shachar (2013)
brand
Foreign (vs. Whether the sponsor brand of an influencer Binary variable equal to 1 if the sponsor brand of an —
local) brand marketing post is a foreign or local brand influencer marketing post is a foreign (vs. local
Chinese) brand
Brand category The category of the sponsor brand of an Binary variable equal to 1 if the sponsor brand of an —
influencer marketing post influencer marketing post belongs to a particular

(continued)
Leung et al. 11

Table 2. (continued)

Variable Definition Operationalization References

category (e.g., beauty products, e-commerce
platforms, electronics)
Influencer Gender of the online influencer Binary variable equal to 1 if the online influencer is Hughes,
gender female (vs. male) Swaminathan, and
Brooks (2019)
Golden week Whether an influencer marketing post is Binary variable equal to 1 if an influencer marketing Ishihara and Ching
initiated during the Chinese Golden Week post is initiated during the Chinese Golden Week (2019)
holiday in October holiday
Weekend Whether an influencer marketing post is Binary variable equal to 1 if an influencer marketing Hughes,
initiated on weekends post is initiated on weekends Swaminathan, and
Brooks (2019)

as related to a new product launch if its content includes one or probit model specified as follows:
more words from the dictionary (= 1, otherwise 0) (for more
coding details, see Web Appendix A1). Pr(Selectionjk = 1|Xj , Zjk ) = Φ(Xj θ + Zjk η), (1)

where j and k denote influencer j and brand k; Selectionjk = 1

Control variables. We added a series of control variables that (selection) or 0 (nonselection); Φ is the cumulative distribution
may influence engagement. At the post level, a dummy indi- function of the standard normal distribution; Xj is a vector of
cates whether an influencer marketing post is related to pro- influencer-level variables that relate directly to an influencer’s
motion (see Web Appendix A1), which is typically for selection, including gender, follower size, and influencer activ-
encouraging consumer trial (Hughes, Swaminathan, and ity; θ and η are vectors of regression coefficients.
Brooks 2019). We also controlled for post length, measured Furthermore, Zjk in Equation 1 includes two control vari-
as the total number of characters in a post. At the brand ables that affect a post’s influencer selection but do not directly
level, we controlled for post number in campaign, or the affect the engagement generated by a post, such that they satisfy
number of posts within an influencer marketing campaign exclusion restrictions (Gill, Sridhar, and Grewal 2017;
launched by the brand. Brand type dummies indicate if it rep- Heckman 1979; Hughes, Swaminathan, and Brooks 2019).
resents a service (vs. product) brand, premium (vs. value) The first control variable is the selection of the second4 most
brand, and foreign (vs. local) brand (see Web Appendix similar influencer (to a focal influencer j) by a brand k. Using
A2). To control for brand heterogeneity, we noted the fixed the approach suggested by Hughes, Swaminathan, and Brooks
effects of brand category, using 29 categories (e.g., beauty (2019), we created an influencer-by-brand matrix (i.e., adja-
products, e-commerce platforms, food and beverages, elec- cency matrix reflecting the selection of influencer j by brand
tronics, apparel, personal care). We also controlled for the k) and multiplied it by its transpose to derive an influencer-
effects of a holiday (i.e., Chinese Golden Week holiday in by-influencer matrix, whose off-diagonal elements indicate
October) and weekends. Finally, we noted the influencer’s influencers who coappeared (i.e., selected by the same brand)
gender. Alternative model specifications also allow for day- most frequently. We then selected the influencer that is the
fixed effect and for influencer-level variation in both the second most similar to (i.e., coappears the second most fre-
intercept and slopes. quently with) a focal influencer j and used the selection (yes =
1, or 0 otherwise) of this “second most similar” influencer by
brand k as a control variable in Zjk. This control variable fulfills
Model Specification and Results the relevance criterion (correlates with the selection of the focal
Selection Model influencer) because in practice, brands tend to recruit influ-
encers with similar characteristics (Hughes, Swaminathan, and
To obtain accurate estimates of the effect of influencer market- Brooks 2019). It also meets the exclusion restriction because
ing spend on engagement (i.e., number of reposts),3 we first even a similar influencer cannot directly affect second-stage
addressed the potential for selection bias in our data. Certain engagement prompted by a post initiated by another (the
influencers, because of their distinct characteristics, may be focal) influencer. The second control variable in Zjk refers to
intentionally selected by brands to engage consumers. increasing (or decreasing) trends in influencer activity.
Therefore, we implemented a Heckman (1979) selection
model to predict an influencer’s selection, with a first-stage
4
We selected the second most similar (instead of the most similar) influencer to
mitigate the possibility that the most similar influencer shares a large proportion
3
The main dependent variable is the number of reposts that an influencer mar- of followers with the focal influencer, whose followers may have seen a compa-
keting post generates within 24 hours. For a robustness check, we also consider rable post initiated by the most similar influencer and therefore have lower
the number of comments a post generates within 24 hours. engagement with the focal influencer’s post.
Table 3. Descriptive Statistics and Correlations.

Correlation Matrix

Variables Mean Median SD Min Max 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

1. Engagement 1,037.78 201.00 21,060.69 .00 1,582,398.00 1.00

2. Influencer 1,002.97 292.86 2,622.79 .43 92,857.14 .33*** 1.00
marketing spend
(in dollars)
3. Influencer 13.37 4.93 19.90 .00 187.67 −.02. −.12*** 1.00
activity
4. Influencer .61 .65 .21 .03 1.00 −.01 −.09*** .33*** 1.00
originality
5. Follower size (in 4.36 2.54 6.70 .00 81.07 .02· .47*** .13*** .04** 1.00
millions)
6. Follower–brand .08 .07 .09 .00 .50 −.01 .01 −.17*** −.12*** −.11*** 1.00
fit
7. Post positivity .28 .31 .42 −.90 .95 .02 −.01 −.10*** −.06*** −.07*** .06*** 1.00
8. Sponsor salience .61 .00 .86 .00 10.00 .07*** .09*** −.15*** −.06*** .05*** .00 .05*** 1.00
9. New product .10 .00 .30 .00 1.00 .00 .04** −.12*** −.05*** −.05*** .13*** −.02· .05*** 1.00
launch
10. Promotion .18 .00 .38 .00 1.00 .04*** .07*** −.14*** −.08*** .02· .00 .12*** .28*** −.02· 1.00
11. Post length 96.58 100.00 48.72 4.00 576.00 .04** .17*** −.43*** −.27*** −.05*** .24*** .08*** .36*** .19*** .27*** 1.00
12. Post number in 47.44 14.00 70.88 1.00 265.00 −.01 −.12*** .24*** .11*** .00 −.10*** .16*** −.07*** −.11*** −.04** −.36*** 1.00
campaign
13. Service (vs. .21 .00 .41 .00 1.00 .03** .11*** −.03* −.08*** .16*** −.17*** −.06*** .14*** −.15*** .16*** .07*** −.06*** 1.00
product) brand
14. Premium (vs. .12 .00 .33 .00 1.00 .00 .04** −.14*** −.07*** −.09*** .10*** .04** −.05*** .14*** −.02 .18*** −.14*** −.09*** 1.00
value) brand
15. Foreign (vs. .34 .00 .47 .00 1.00 .00 .03* −.19*** −.08*** −.11*** .10*** .02· .02 .09*** .05*** .28*** −.08*** −.33*** .36*** 1.00
local) brand
16. Influencer .64 1.00 .48 .00 1.00 −.01 −.02† −.10*** −.08*** −.08*** .10*** .02 −.07*** −.01 .02 .08*** −.03* −.12*** .05*** .16*** 1.00
gender
17. Golden week .08 .00 .27 .00 1.00 −.01 −.07*** .17*** .10*** .01 −.03** −.03* −.08*** −.08*** −.09*** −.18*** .07*** −.06*** −.05*** −.09*** .00 1.00
18. Weekend .26 .00 .44 .00 1.00 .02 −.01 .00 −.02 −.01 .01 −.01 .00 −.02 .04*** −.03* .01 .00 −.02 −.01 −.02· .10*** 1.00

†
p < .10, *p < .05, **p < .01, ***p < .001,.
Notes: The statistics are reported based on the original values of variables (i.e., without taking natural logarithms).
Leung et al. 13

Specifically, we calculated the time trend5 of each influencer on Table 4. First-Stage Selection Model.
the basis of posting activities in the preceding 90 days; this
Influencer Selection
control variable satisfies the exclusion restriction. A brand Independent Variables (1 = Selection)
manager may scan the historical posts of an influencer when
making the recruitment decision, but second-stage engagement Similar influencer selection (1 = selection) 1.646***
prompted by a focal post is unlikely to be influenced by prior (.018)
posts. Increasing trend in influencer activity .030
Table 4 contains the estimation results of the first-stage selec- (.024)
tion model. In line with our predictions, a focal influencer is sig- Influencer gender (1 = female) .069***
nificantly more likely to be selected by a brand when a similar (.010)
Influencer activity <.001
influencer is also selected (b = 1.646, z = 89.85, p < .001).
(<.001)
Selection probabilities are also significantly greater for female
Follower size .011***
influencers (b = .069, z = 6.78, p < .001) and those with larger
(<.001)
follower sizes (b = .011, z = 14.87, p < .001). Higher influencer
Constant −2.980***
activity (b < .001, z = 1.02, p = .309) and the increasing trend (.009)
in their activity (b = .030, z = 1.25, p = .212), though not sig- Number of observations 2,076,732
nificant, are associated with larger selection probabilities.
The inverse Mills ratio (IMR), representing unobserved ***p < .001.
aspects of influencer selection determined by brands and influ- Notes: Standard errors are reported in parentheses.
encers (e.g., prior relationship between them), is included as an
independent variable in all second-stage models, alleviating
the concern that the estimated effect of influencer marketing where i and j denote post i and influencer j; ln(Engagementij )
spend on engagement is prone to selection bias. and ln(IMSij ) are the natural logarithms of engagement (i.e.,
number of reposts) and of influencer marketing spend,
respectively;6 and the seven moderating variables are as
Engagement Model defined in the hypothesized conceptual model. Note that
the interaction terms involve variables that are not mean cen-
To estimate the effect of influencer marketing spend on engage- tered. Furthermore, Cij is vector of control variables, includ-
ment, we used a log-log specification for the second-stage ing promotion, post length, post number in campaign, brand
model to facilitate the interpretation of the coefficients for types (service vs. product brand, premium vs. value brand,
(log) influencer marketing spend as engagement elasticity, foreign vs. local brand), brand categories (29 categories),
which is a unit-free and generalizable measurement. By identi- influencer gender, Golden Week, weekend, and the IMR
fying factors that moderate this elasticity, we also aim to under- from the first-stage model. Finally, β0, β1, …, β15, and α
stand what enhances (or weakens) the effectiveness of are (scalar or vector) coefficients, and εij denotes an indepen-
influencer marketing spend. To that end, we estimate the follow- dent error term.
ing second-stage engagement model: We started the analysis with lower-order terms (i.e., influ-
ln(Engagementij ) = β0 + β1 ln(IMSij ) + β2 Influencer Activityij encer marketing spend, seven moderators, and controls)
(Model 1). The results in Table 5 indicate substantive, signifi-
+ β3 Influencer Originalityij + β4 Follower Sizeij cant engagement elasticity; ceteris paribus, a 1% increase in
influencer marketing spend increases engagement by .457% (t
+ β5 Follower Brand Fitij + β6 Post Positivityij
= 16.96, p < .001). Therefore, increasing the influencer market-
+ β7 Sponsor Salienceij + β8 New Product Launchij ing budget can increase engagement. Note that the variance
inflation factors in Model 1 (no interaction terms) are all
+ β9 Influencer Activityij × ln(IMSij )
below 2.0, alleviating collinearity concerns.
+ β10 Influencer Originalityij × ln(IMSij ) We then included the interaction terms, as specified in the
engagement model in Equation 2 (Model 2), to test H2, H3,
+ β11 Follower Sizeij × ln(IMSij ) + β12 Follower Brand Fitij H6, and H7. We find that influencer originality significantly
× ln(IMSij ) + β13 Post Positivityij × ln (IMSij ) enhances engagement elasticity (b = .592, t = 5.18, p < .001),
+ β14 Sponsor Salienceij × ln (IMSij ) such that engagement is more responsive to greater influencer
marketing spend when the influencer posts more original
+ β15 New Product Launchij × ln (IMSij ) + Cij α + εij , content, in support of H2. In addition, as hypothesized in H3,
(2) engagement elasticity increases significantly with larger

5 6
To obtain the time trend, we regressed an influencer’s posting activity on each We added .01 to the number of reposts before taking the logarithm to include
day against the days in chronological order (t = 1, 2, …, 90). posts that were not reposted.
14 Journal of Marketing 0(0)

Table 5. Hypothesis Tests: Second-Stage Engagement Model and Nonlinear Moderating Effects.

Variables Model 1 Model 2 Model 3a Model 3b Model 3c Model 3

Independent Variable
ln(Influencer marketing spend) ($IMS) .457*** .143† .076 .098 .183* .072
(.027) (.080) (.081) (.081) (.082) (.084)
Interaction Terms
Sender (Influencer) Characteristics
Influencer activity × ln($IMS) .006*** .020*** .006*** .006*** .020***
(.002) (.003) (.002) (.002) (.003)
Influencer activity2 × ln($IMS) H1 (∩) −.001*** −.001***
(<.001) (< .001)
Influencer originality × ln($IMS) H2 (+) .592*** .561*** .594*** .593*** .565***
(.114) (.113) (.114) (.114) (.113)
Follower size × ln($IMS) H3 (+) .019*** .010** .019*** .019*** .010**
(.003) (.003) (.003) (.003) (.003)
Receiver (Follower) Characteristics
Follower‐brand fit × ln($IMS) −.431 −.395 1.411* −.444† 1.358*
(.268) (.265) (.652) (.268) (.644)
Follower‐brand fit2 × ln($IMS) H4 (∩) −8.185** −7.825**
(2.647) (2.612)
Message (Post) Characteristics
Post positivity × ln($IMS) −.087† −.064 −.087† −.007 .017
(.049) (.048) (.049) (.060) (.059)
Post positivity2 × ln($IMS) H5 (∩) −.250* −.250*
(.108) (.107)
Sponsor salience × ln($IMS) H6 (+/−) .092*** .100*** .097*** .087** .100***
(.028) (.027) (.028) (.028) (.027)
New product launch × ln($IMS) H7 (+/−) −.488*** −.441*** −.501*** −.475*** −.440***
(.070) (.069) (.070) (.070) (.070)
Moderators
Influencer activity −.036*** −.055*** −.177*** −.055*** −.055*** −.177***
(.002) (.007) (.015) (.007) (.007) (.015)
Influencer activity2 .001*** .001***
(<.001) (<.001)
Influencer originality −2.043*** −5.213*** −4.991*** −5.224*** −5.212*** −5.002***
(.184) (.682) (.675) (.681) (.682) (.675)
Follower size .010 −.159*** −.080** −.161*** −.159*** −.081**
(.006) (.028) (.029) (.028) (.028) (.029)
Follower‐brand fit .020 2.051 1.865 −8.579* 2.134 −7.893*
(.451) (1.625) (1.606) (3.980) (1.625) (3.930)
Follower‐brand fit2 47.624** 44.091**
(16.268) (16.058)
Post positivity .013 .463 .294 .459 .052 −.118
(.090) (.290) (.287) (.290) (.354) (.349)
Post positivity2 1.306* 1.299*
(.649) (.641)
Sponsor salience .009 −.561** −.609*** −.588*** −.528** −.599***
(.046) (.178) (.176) (.178) (.179) (.177)
New product launch (1 = yes) .302* 3.224*** 2.907*** 3.337*** 3.139*** 2.925***
(.128) (.445) (.440) (.447) (.447) (.444)
Control Variables
Promotion (1 = yes) .075 .136 .133 .142 .141 .144
(.101) (.100) (.099) (.100) (.100) (.099)
Post length .012*** .011*** .009*** .011*** .011*** .009***
(.001) (.001) (.001) (.001) (.001) (.001)

(continued)
Leung et al. 15

Table 5. (continued)

Variables Model 1 Model 2 Model 3a Model 3b Model 3c Model 3

Post number in campaign −.007*** −.007*** −.006*** −.007*** −.007*** −.006***
(<.001) (<.001) (<.001) (<.001) (<.001) (<.001)
Service brand (1 = yes) .491* .626** .823*** .637** .639** .847***
(.241) (.240) (.237) (.240) (.240) (.237)
Premium brand (1 = yes) .147 −.010 −.031 <.001 −.013 −.029
(.127) (.126) (.125) (.126) (.126) (.125)
Foreign brand (1 = yes) −.059 −.043 −.098 −.052 −.032 −.096
(.101) (.102) (.101) (.102) (.102) (.101)
Influencer gender (1 = female) .158* .205** .277*** .210** .198* .274***
(.078) (.077) (.077) (.077) (.077) (.077)
Golden week (1 = yes) −1.006*** −.901*** −.869*** −.897*** −.896*** −.860***
(.141) (.139) (.138) (.139) (.139) (.138)
Weekend (1 = yes) .048 .042 .024 .034 .043 .019
(.083) (.082) (.081) (.082) (.082) (.081)
IMR <.001 −.016 −.007 −.014 −.018 −.008
(.062) (.061) (.061) (.061) (.061) (.061)
Constant 2.357*** 4.415*** 5.187*** 4.642*** 4.230*** 5.221***
(.329) (.549) (.549) (.554) (.558) (.562)
Brand category-fixed effects Yes Yes Yes Yes Yes Yes
Number of observations 5,835 5,835 5,835 5,835 5,835 5,835
Adjusted R-square .421 .435 .450 .436 .436 .451
AIC 28,171 28,030 27,882 28,024 28,028 27,874
BIC 28,491 28,397 28,262 28,404 28,408 28,281
†
p < .10, *p < .05, **p < .01, ***p < .001.
Notes: Standard errors are reported in parentheses.

follower size (b = .019, t = 5.92, p < .001). We find that engage- variable. The results for H2, H3, H6, and H7 remain
ment elasticity increases when the sponsor brand is more salient consistent and robust, offering further confidence in the
in the post (b = .092, t = 3.33, p < .001), in support of H6a. This findings derived from our engagement model (see Web
finding suggests that instead of prompting ad avoidance, Appendix E).
sponsor salience likely enhances content value by making the
post more informative. Finally, engagement elasticity dimin-
ishes in response to posts about new product launches (b =
Delineating the Nonlinear Moderating Effects
−.488, t = −6.96, p < .001), consistent with H7b. The result sug- To test H1, H4, and H5, we considered the inverted U-shaped
gests that the risks surrounding new product launches likely moderating effects of three moderators—influencer activity, fol-
prohibit consumers’ responsiveness and weaken engagement lower–brand fit, and post positivity—on engagement
elasticity. The coefficients of the IMR are nonsignificant in elasticity. Specifically, we added interactions of the quadratic
Models 1 and 2; unobserved aspects of influencer selection, forms of the proposed moderators with ln(IMSij ) (i.e.,
such as prior relationships between influencers and brands, Influencer Activity2ij × ln(IMSij ), Follower Brand Fit2ij × ln(IMSij ),
thus appear unlikely to affect engagement. For completeness, and Post Positivity2ij × ln(IMSij )) and the simple quadratic
we introduced the interactions separately and provide these forms of the three moderators (i.e., Influencer Activity2ij ,
model estimation results in Web Appendix B, Models 2a–2c. Follower Brand Fit2ij , and Post Positivity2ij ) to the engagement
Because the trend in influencer activity (i.e., one of the variables model in Equation 2 to test for curvilinear moderating effects
used in the exclusion restriction) is nonsignificant in the selec- (Jaccard and Turrisi 2003; Lind and Mehlum 2010; Wielgos,
tion model, we alternatively estimated the engagement model Homburg, and Kuehnl 2021). We introduced the interactions of
by excluding this variable from the selection model (see Web each quadratic moderator with ln(IMSij ) stepwise (Models 3a–3c),
Appendix C). Furthermore, Web Appendix D contains the esti- then collectively (Model 3).
mation results without the IMR. All the estimates remain similar As Table 5 reveals, the squared interaction term between
in sign, relative magnitude, and significance. influencer activity and ln(IMSi ) is negative and significant
As a robustness check, we used another engagement (Model 3a: b = −.001, t = −4.19, p < .001; Model 3: b =
metric—number of comments an influencer marketing −.001, t = −4.21, p < .001), indicating that engagement elastic-
post generates within 24 hours of posting—as the dependent ity varies across different levels of influencer activity in an
16 Journal of Marketing 0(0)

Figure 2. Nonlinear moderating effects on influencer marketing elasticity.

Notes: TP = turning point.

inverted U-shaped manner, in support of H1. Similarly, the moderating effects. As Panel A shows, engagement elasticity
results reveal significant inverted U-shaped moderating effects increases when influencer activity rises to an optimal level, then
for follower–brand fit (Model 3b: b = −8.185, t = −3.09, p < diminishes beyond that turning point (TPinfluencer activity =
.01; Model 3: b = −7.825, t = −3.00, p < .01) and post positivity 64.781). We note a significant positive slope at the low end of
(Model 3c: b = −.250, t = −2.32, p < .05; Model 3: b = −.250, t influencer activity (b = .020, t = 6.71, p < .001) and a negative
= −2.35, p < .05), in support of H4 and H5, respectively.7 The slope at the high end (b = −.038, t = −3.36, p < .001).8 Similar
moderating effects of influencer originality, follower size, inverted U-shaped patterns appear in Figure 2, Panels B and C,
sponsor salience, and new product launch remain robust to revealing that follower–brand fit and post positivity strengthen
these alternative model specifications (Models 3a–3c and engagement elasticity initially (follower–brand fit: bleft = 1.358, t
Model 3), such that we obtain strong evidence for the validity = 2.11, p < .05; post positivity: bleft = .466, t = 2.02, p < .05), then
of H2, H3, H6, and H7. weaken engagement elasticity when they reach high levels (fol-
To probe these results further, we attempted to predict engage- lower–brand fit: bright = −6.439, t = −3.17, p = < .001; post
ment elasticity conditional on the values of influencer activity, fol- positivity: bright = −.457, t = −2.61, p < .01). The turning
lower–brand fit, and post positivity (Model 3; all other variables points are located well within the respective data ranges
remained at their mean) and plotted the predicted values. (TPfollower–brand fit = .087; TPpost positivity = .033). These
Figure 2 illustrates the hypothesized inverted U-shaped results provide further evidence to support H1, H4, and H5.

7 8
We experimented with specifications that also included cubic interactions of We calculated the slopes at low and high values of the moderators, as well as
the three moderators with ln(IMSij ); their nonsignificant coefficients indicate their turning points, using unstandardized regression coefficients (Jaccard and
support for inverted U-shaped moderating effects. Turrisi 2003; Wielgos, Homburg, and Kuehnl 2021).
Leung et al. 17

To confirm the robustness of our findings, we employed another graphically depicts the inverted U-shaped moderating effects
approach to investigate the nonlinear moderating effects. of post positivity, yielded by these two algorithms, on engage-
Specifically, for each proposed moderator, we divided posts with ment elasticity, providing visual evidence in support of H5.
a median split, then estimated Equation 2 separately for the two sub- Finally, we checked the level of analysis. Both the dependent
samples. If inverted U-shaped relationships exist between the mod- variable (i.e., number of reposts) and the focal independent var-
erators and engagement elasticity, we should observe initially iable (i.e., influencer marketing spend) are at the influencer mar-
strong elasticity (low levels of the moderators) that diminishes as keting post level, so single posts provide the fundamental unit of
the moderators increase (high levels of the moderators). Web analysis. Our empirical analyses based on each post identify the
Appendix F presents the estimation results. Consistent with H1, impact of post-level heterogeneity on engagement elasticity. But
engagement elasticity strengthens at first (b = .091, t = 7.08, p < our data also contain information at the influencer level, such that
.001) but weakens (b = −.003, t = −1.58, p = .114) with increasing we can nest the individual posts within different influencers. To
influencer activity (Web Appendix F, Columns 1 and 2). Higher test the validity of our hypotheses, we therefore implemented a
follower–brand fit also first increases (b = 3.752, t = 2.08, p < mixed-effects model that controls for potential heterogeneity
.05), then decreases (b = −1.961, t = −4.19, p < .001) engagement among influencers, using the following alternative specification:
elasticity, in support of H4 (Web Appendix F, Columns 3 and 4). In
a similar nonlinear pattern, a higher level of post positivity initially ln(Engagementij ) = b0j + b1j ln(IMSij ) + b2j Influencer Activityij
strengthens elasticity (b = .206, t = 1.82, p = .069), then weakens it
+ b3j Influencer Originalityij + b4j Follower Sizeij
(b = −.191, t = −1.10, p = .272), in line with H5 (Web Appendix F,
Columns 5 and 6). This median-split approach provides managerial + b5j Follower Brand Fitij + b6j Post Positivityij
insights related to the two subsamples: Brands can use their industry
medians as a benchmark, then decide whether to increase or + b7j Sponsor Salienceij + b8j New Product Launchij
decrease the levels of the moderators to achieve more effective + b9j Influencer Activity2ij + b10j Follower Brand Fit2ij
influencer marketing campaigns.
+ b11j Post Positivity2ij + b12j Influencer Activityij × ln(IMSij )
Additional Robustness Check + b13j Influencer Activity2ij × ln(IMSij )
We already replicated our main model results with an alternative + b14j Influencer Originalityij × ln (IMSij )
dependent variable (i.e., number of comments) and validated the
nonlinear moderating effects with different approaches to ensure + b15j Follower Sizeij × ln(IMSij )
the robustness of our findings. In this section, we present several + b16j Follower Brand Fitij × ln(IMSi )
additional robustness checks to provide further evidence of the
validity of the proposed effects. Web Appendix G, Column 1, + b17j Follower Brand Fit2ij × ln(IMSij )
lists the results when we took the natural logarithm of follower
+ b18j Post Positivityij × ln(IMSij ) + b19j Post Positivity2ij
size to account for its skewness. In Column 2, we present the
Model 3 (full model) results when we controlled for the fixed × ln(IMSij ) + b20j Sponsor Salienceij × ln(IMSij )
effects of days (i.e., every day of October) to alleviate concerns
+ b21j New Product Launchij × ln(IMSij ) + Cij d + ϵij ,
of seasonality in consumer engagement. Column 3 depicts the
results when we operationalized sponsor salience with only the (3)
count of URLs linked to the sponsor brand’s website or online
shop. In all columns, the estimation results for H1–H7 are consistent where i and j denote post i and influencer j; most of the variables
with those obtained from our initial Model 3, suggesting the pro- are as previously defined; b0j is the random intercept; and
posed effects are robust to the alternative model specifications. b1j , b2j , . . . , b21j are random slopes. We denote
Regarding the moderating role of post positivity, we com- bj = (b0j , b1j , b2j , . . . , b21j )′ , such that bj is multivariate nor-
plement our initial deductive, top-down dictionary-based mally distributed bj ∼ MVN(μ, G). We also assume the
approach with abductive, bottom-up approaches to offer con- random effects are independent;9 the off-diagonal elements of
vergent insights (Humphreys and Wang 2018). Specifically, the variance covariance matrix G are 0. Finally, d is a vector of
we used two proven, reliable algorithms for text classification: fixed coefficients, and ϵij is an independent error. We first
a naive Bayesian classifier and a support vector machine clas- included the random intercept (Web Appendix I, Column 1),
sifier (Tirunillai and Tellis 2012). Then, we reestimated Model then all the random effects specified in Equation 3 (Web
3 using the post positivity yielded by the two algorithms. Appendix I, Column 2). The results are comparable to those in
The estimation results are consistent with H5; as we specify Table 5, Model 3, revealing consistent estimates across the three
in Web Appendix H, the squared interaction term between models in their signs, relative magnitude, and significance.
post positivity and ln(IMSij ) is significantly negative for both
algorithms (naive Bayesian classifier: b = −1.998, t = −2.31, 9
Estimating the dependence between influencer-level random effects is theoret-
p < .05; support vector machine classifier: b = −1.358, t = ically feasible but computationally intractable with our empirical data, which
−3.14, p < .01; Table WH1, Columns 1 and 2). Figure WH1 contain many random effects (i.e., 2,412 influencers).
18 Journal of Marketing 0(0)

Thus, the results appear robust to alternative specifications that suboptimally allocating their budgets across influencers, and
account for influencer-level mixed effects. there is considerable upward potential in generating engage-
ment if they were to account for influencers’ engagement elas-
ticities (and base engagement levels) in their decisions.
Post Hoc Analyses for Managerial Insights In another set of auxiliary analyses, we varied each moderat-
ing factor and examined their impacts on influencer marketing
An advantage of estimating the engagement model is that we can effectiveness (i.e., engagement elasticity), while keeping all
derive practical insights into how firms can better allocate their other variables at their mean levels (see Web Appendix J).
influencer marketing budget across influencers and help firms The results offer clearer insights into the managerially relevant
understand how much upward potential they have. We take the magnitude of these moderating effects. They also highlight
perspective of firms that incentivize influencers to maximize some unique tensions for marketers who design influencer mar-
profits through engaging followers. Assuming a constant engage- keting campaigns. When selecting on the basis of influencer
ment elasticity for each influencer (i.e., for a given influencer the activity, follower–brand fit, and post positivity, more is not
elasticity is the same across the range of influencer marketing always better. For example, a sponsored post shared by a
spend, in line with the log-log model that was used), the key deci- highly active influencer, one with high levels of follower–
sion is how to allocate influencer marketing budget across influ- brand fit, and one showing high post positivity (2 SD beyond
encers, regardless of the number of posts the allocated amount the optimal points) would lose 21.0%, 31.4%, and 22.4% in
translates into. We focus on the budget allocation decision, engagement elasticity, respectively.
rather than on determining the size of the budget. Specifically,
given firms’ existing influencer marketing budget, we compare
how firms should allocate their budget across influencers opti- General Discussion
mally versus how they actually have allocated their budget. This research sheds new light on the factors that alter influencer
We leverage the optimal allocation rule recommended by marketing effectiveness. By accounting for influencer market-
Peers, Van Heerde, and Dekimpe (2017). Assuming equal per- ing spend, we assess the efficacy of leveraging influencers to
engagement profit contribution for all influencers, the optimal encourage consumer engagement. Factors related to the influ-
budget allocation is proportional to the (product of) engagement encers, their followers, and their sponsored posts all have criti-
elasticity and the base engagement levels of the influencers. The cal roles in determining the level of influencer marketing
intuition is that more budget should be allocated to those with effectiveness. In outlining these nuanced effects, we reveal
stronger responsiveness to sponsorship and larger size of base the importance of communication factors for decisions about
engagement. Consider an example from our data set: A hair influencer marketing campaigns.
styling brand spent $264 on influencer marketing, including
$234 (89%) for posts by Influencer A and $30 (11%) for
posts by Influencer B. Their engagement elasticities are .16 Theoretical Contributions
and .57, while their base engagement levels are 19 and 15 This research enriches existing marketing literature by integrat-
reposts, respectively. This means the budget allocation ing firms’ spending on influencer marketing as a measure of
weights should have been 3.04 (i.e., .16 × 19) for Influencer A influencer marketing effectiveness. Prior studies of influencer
and 8.55 (i.e., .57 × 15) for Influencer B or 26.2% (i.e., marketing mostly document the impacts of source and post
[.16 × 19]/[.16 × 19 + .57 × 15]) for Influencer A and 73.8% characteristics on various key outcomes (see Table 1) but lack
(i.e., [.57 × 15]/[.16 × 19 + .57 × 15]) for Influencer B. If this critical assessments of the costs required by campaigns.
firm had adopted this optimal budget allocation, it could have Because influencer spend varies substantially, ignoring such
increased consumer engagement by 42.0%, which corresponds costs limits the accuracy of the findings regarding the effective-
to 163 more reposts (from 389 to 552 reposts), compared with ness of influencer marketing to engage consumers (Batra and
consumer engagement achieved with its actual allocation. Keller 2016). We extend efforts to integrate media spending
When we perform similar calculations for all firms in our data in determinations of advertising elasticities (Dinner, Van
set, to determine the difference in engagement achieved with Heerde, and Neslin 2014; Sethuraman, Tellis, and Briesch
optimal versus actual budget allocations, we find that on 2011) to the domain of influencer marketing. As research into
average, the firms could achieve a 16.6% increase in engage- influencer marketing grows, it is critical to investigate the
ment.10 This finding suggests that the firms in our data set are extent of its effectiveness by using data about both cost and
engagement outcomes because achieving higher engagement
10
Influencers’ engagement elasticities and base engagement levels were returns remains a key concern for marketers.
obtained through estimating Equation 3 (mixed-effects model) with all indepen- Drawing on communication theory (Lasswell 1948;
dent variables mean-centered. Alternatively, we measured influencers’ base Shannon and Weaver 1949), this study delineates how
engagement levels with the average number of reposts across their posts, and diverse factors pertaining to the influencer (sender), followers
the optimal allocation results are similar, with a 16.7% increase in engagement.
Budget allocations across influencers require the existence of more than one (receiver), and posts (message) can influence influencer mar-
influencer, so this analysis refers only to firms that contract with more than keting effectiveness. In relation to sender characteristics, we
one influencer. identify three easily observable characteristics (influencer
Leung et al. 19

activity, originality, and follower size). Influencer activity widely. Without cost considerations, firms cannot accurately
exerts an inverted U-shaped effect; excessive activity evaluate the effectiveness of their influencer marketing efforts.
appears to lead to information overload. This finding thus In establishing a positive effect of influencer marketing spend
helps reconcile mixed results in prior studies pertaining to on consumer engagement, our findings confirm the effective-
the effects of posting activity (Stephen et al. 2017; Suh ness of this communication practice. The estimation of engage-
et al. 2010). Moreover, unlike celebrity endorsers, influencers ment elasticity also sheds light on how firms actually allocate
can create content for brands, and the extent to which they their budgets across influencers compared with how they
share self-created (vs. other-created) content offers a unique should do so. As our post hoc analysis indicates, the average
characteristic for consideration. We find that influencers firms in our data set are allocating their budgets suboptimally
who post more original content are more effective senders and have substantive upward potential regarding generating
because they increase engagement elasticity. In line with engagement.
seeding strategy research, which generally suggests the supe- Our findings also offer insights for managing influencer mar-
riority of seeding well-connected members of a network (Hinz keting, in terms of influencer selection, follower considerations,
et al. 2011; Libai, Muller, and Peres 2013), we find that fol- and content briefings. In addition to investing in influencers
lower size enhances effectiveness. with more followers, firms can enhance elasticity by selecting
Characteristics of followers also define a campaign’s
influencers who transmit more original content. Furthermore,
effectiveness. Extant endorsement literature often cites a pos-
itive impact of brand fit (Bergkvist and Zhou 2016), but as we they should encourage influencers to make the sponsor brand
find, follower–brand fit can have an inverted U-shaped effect more salient in the posts by incorporating clickable brand men-
on influencer marketing effectiveness. This result echoes findings tions and URL links. However, by accounting for influencer
in behavioral targeting literature (Boerman, Kruikemeier, and marketing spend, our findings may appear to contradict some
Borgesius 2017) that suggest both the benefits and downsides managerial intuitions. For example, posts introducing new prod-
of precise targeting approaches. But as long as follower–brand
ucts may be shared more because they contain novel and inter-
fit is not too high, influencer marketing can deliver surprises to fol-
esting content (Tellis et al. 2019), but firms should also be aware
lowers, which is an important goal of marketers, considering the
mounting advertising clutter. that their engagement elasticity is lower than that of posts that
Finally, our investigation of post characteristics demon- remind consumers of existing products, likely because of the
strates some of the unique features of influencer marketing. heightened risks associated with new product launches. Yet
Previous literature suggests that more positive content is more the firms in our data set spent significantly more on influencer
viral (Berger and Milkman 2012), but we find that a blend of marketing posts that promote new products. Existing products
positive and negative content can yield the greatest effective-
receive relatively less attention and resources, although such
ness. In most cases, consumers realize that influencers are
paid to advocate for a brand or product, so consumers may allocations would be more effective—consistent with the
start to question the influencers’ authenticity if they only notion that spending on less favored situations is beneficial
share positive comments about an endorsed brand (Van Heerde et al. 2013).
(McQuarrie, Miller, and Phillips 2013). Moreover, previous lit- Firms also must attend to three key tensions related to influ-
erature offers conflicting views regarding the impact of sponsor encer activity, follower–brand fit, and post positivity. Efforts to
salience on advertising effectiveness. Our results support the move these decision factors toward their optimal levels can
view that including more links to the sponsor brand can greatly enhance the effectiveness of influencer marketing
enhance a post’s informativeness (Lou and Yuan 2019), con- spend. Specifically, influencers who engage in a medium level
firming that sponsor salience enhances influencer marketing of posting activity are preferable; in Figure 2, Panel A, we
effectiveness. Last but not least, although there are competing find that average firms in our data set could have increased
arguments regarding whether new product launches motivate the effectiveness of their influencer marketing efforts by
or inhibit consumers’ tendency to engage with influencer mar- 53.8% if they selected influencers who engaged in more
keting efforts, our results confirm the latter view in that consum- posting activity, up to the optimal level. However, followers
ers are less responsive to posts about new product launches, with a high degree of shared interest with the brand may not
likely because of heightened risks and social desirability be the best group to target. The average firms in our data set
concerns. were already targeting follower groups at nearly optimal
levels of follower–brand fit (Figure 2, Panel B). Finally, firms
should encourage influencers to include some critical content
Managerial Implications in their posts to enhance the perceived credibility of the mes-
Influencer marketing enables firms to crowdsource the sages. Figure 2, Panel C, indicates that the average firms in
resources that influencers possess, including their follower net- our data set overshoot in terms of post positivity; they could
works, creative content, personal positioning, and follower increase their influencer marketing effectiveness by 1.9% by
trust, to enhance marketing communications (Leung, Gu, and instructing influencers to offer some slightly negative content
Palmatier 2022). However, the costs of these campaigns vary in their posts.
20 Journal of Marketing 0(0)

claims may affect influencers’ communication with followers.

Further Research and Limitations In addition, our data set consists of influencer marketing posts
Our study is subject to several limitations that also suggest gathered over one month; data spanning a longer period, involv-
future research directions. First, we focus on influencer activity, ing more influencers (especially smaller ones), could help gener-
originality, and follower size as important drivers of influencer alize the insights from the current study.
marketing effectiveness. Research that assesses other influencer Fourth, even if generating consumer engagement (e.g., likes,
characteristics, such as consistency (based on historical posts), comments, reposts) is a primary objective of influencer market-
personality (e.g., humorous, confident), perceived uniqueness, ing campaigns, not every form of engagement is created equal.
authenticity, and endorsement history, could help firms gain Consumers may exhibit varied search behavior or purchase con-
more insights into the determinants of effectiveness. Similarly, version likelihoods when they engage with different influencers
we measure follower–brand fit as an important follower charac- or posts. With currently available technologies, firms can track
teristic; additional research could examine followers’ homogeneity click-throughs and website traffic (Lee, Hosanagar, and Nair
(e.g., demographics, interests) or influencer–follower relationship 2018), then measure which sales result from particular influ-
strength to glean more similar insights. encer marketing posts. Further research should examine the
Second, our empirical context is Weibo, a microblogging drivers of click-through– or sales-based elasticity outcomes to
platform similar to Twitter. In operationalizing post characteris- determine how specific influencer marketing strategies could
tics, such as positivity and sponsor salience, we rely on text exert direct impacts on firms’ bottom lines.
analyses. The increasingly diverse platforms and tools available
for influencer marketing may require different content measures
Associate Editor
though. For example, Instagram supports visually engaging
images and YouTube enables influencers to create videos Koen Pauwels
with in-depth information that can remain accessible for a
long time. Content from short-video-sharing apps (e.g., Declaration of Conflicting Interests
TikTok) provide a more short-lived but entertaining function. The author(s) declared no potential conflicts of interest with respect to
Further research could move beyond text-based analyses to the research, authorship, and/or publication of this article.
address content factors such as aesthetics or the entertainment
value of visual content. In addition, our results reveal that
posts about new product launches reduce influencer marketing Funding
effectiveness, possibly because of the heightened risks associ- The author(s) disclosed receipt of the following financial support for
ated with new products. From another view, because firms the research, authorship, and/or publication of this article: This work
tend to overinvest in advertising to promote new products was supported by the Lingnan University Direct Grant, Lingnan
(Krishnan and Jain 2006), repetition and wearout may occur University Faculty Research Grant (grant number DR21B3, DB21A7).
(Bruce, Foutz, and Kolsarici 2012), which diminishes consum-
ers’ responsiveness to specific influencer marketing efforts. ORCID iDs
Although we do not have data to evaluate whether influencer Fine F. Leung https://orcid.org/0000-0002-7988-2944
marketing effectiveness decreases with higher noninfluencer Flora F. Gu https://orcid.org/0000-0002-0560-2014
marketing spend, future research could investigate such poten-
tial negative synergy.
Third, classic models of communication (Lasswell 1948) References
often include details about who (sender) says what (message) Akpinar, Ezgi and Jonah Berger (2017), “Valuable Virality,” Journal
in which channel (medium) and to whom (receiver). We iden- of Marketing Research, 54 (2), 318–30.
tify the effects of sender, receiver, and message characteristics, Balducci, Bitty and Detelina Marinova (2018), “Unstructured Data in
but our data are limited to a single social media platform Marketing,” Journal of the Academy of Marketing Science, 46
(medium), which may have its own unique restrictions. For (4), 557–90.
example, our findings reveal that selecting influencers who Barasch, Alixandra and Jonah Berger (2014), “Broadcasting and
create original posts enhances engagement elasticity but has a Narrowcasting: How Audience Size Affects What People Share,”
negative main effect on consumers’ reposting. This result may Journal of Marketing Research, 51 (3), 286–99.
be specific to Weibo, considering the relatively high levels of Barker, Shane (2018), “4 Mistakes Influencers Should Not Make on
content moderation and control it imposes and the associated Instagram,” Influence (June 10), https://influence.bloglovin.com/4-
risks of reposting for consumers. Additional research could test mistakes-influencers-should-not-make-on-instagram-fdab32e4403e.
our conceptual framework in another medium. Then a related Batra, Rajeev and Kevin Lane Keller (2016), “Integrating Marketing
research direction could address specific channel or medium char- Communications: New Findings, New Lessons and New Ideas,”
acteristics and their impacts on influencer marketing effectiveness. Journal of Marketing, 80 (6), 122–45.
As Shannon and Weaver (1949) suggest, noise in the communica- Becker, Maren, Nico Wiegand, and Werner J. Reinartz (2019), “Does It
tion process may interfere with the channel or distort the message; Pay to Be Real? Understanding Authenticity in TV Advertising,”
noise such as clutter due to competitive influencer posts and Journal of Marketing, 83 (1), 24–50.
Leung et al. 21

Berger, Jonah (2014), “Word of Mouth and Interpersonal De Veirman, Marijke, Veroline Cauberghe, and Liselot Hudders
Communication: A Review and Directions for Future Research,” (2017), “Marketing Through Instagram Influencers: The Impact
Journal of Consumer Psychology, 24 (4), 586–607. of Number of Followers and Product Divergence on Brand
Berger, Jonah and Katherine L. Milkman (2012), “What Makes Online Attitude,” International Journal of Advertising, 36 (5), 798–828.
Content Viral?” Journal of Marketing Research, 49 (2), 192–205. De Veirman, Marijke and Liselot Hudders (2020), “Disclosing
Berger, Jonah and Eric M. Schwartz (2011), “What Drives Immediate Sponsored Instagram Posts: The Role of Material Connection
and Ongoing Word of Mouth?” Journal of Marketing Research, 48 with the Brand and Message-Sidedness When Disclosing Covert
(5), 869–80. Advertising,” International Journal of Advertising, 39 (1), 94–130.
Bergkvist, Lars and Kris Qiang Zhou (2016), “Celebrity Endorsements: Dinner, Isaac M., Harald J. van Heerde, and Scott A. Neslin (2014),
A Literature Review and Research Agenda,” International Journal “Driving Online and Offline Sales: The Cross-Channel Effects of
of Advertising, 35 (4), 642–63. Traditional, Online Display, and Paid Search Advertising,”
Bharadwaj, Neeraj, Michel Ballings, Prasad A. Naik, Miller Moore, Journal of Marketing Research, 51 (5), 527–45.
and Mustafa Murat Arat (2022), “A New Livestream Retail Ducoffe, Robert H. (1996), “Advertising Value and Advertising on the
Analytics Framework to Assess the Sales Impact of Emotional Web,” Journal of Advertising Research, 36 (5), 21–35.
Displays,” Journal of Marketing, 86 (1), 27–47. Eisend, Martin and Farid Tarrahi (2016), “The Effectiveness of
Boerman, Sophie C., Sanne Kruikemeier, and Frederik Advertising: A Meta-Meta-Analysis of Advertising Inputs and
J. Zuiderveen Borgesius (2017), “Online Behavioral Advertising: A Outcomes,” Journal of Advertising, 45 (4), 519–31.
Literature Review and Research Agenda,” Journal of Advertising, 46 Escalas, Jennifer Edson and James R. Bettman (2017), “Connecting
(3), 363–76. with Celebrities: How Consumers Appropriate Celebrity
Boerman, Sophie C., Eva A. Van Reijmersdal, and Peter C. Neijens Meanings for a Sense of Belonging,” Journal of Advertising, 46
(2015), “How Audience and Disclosure Characteristics Influence (2), 297–308.
Memory of Sponsorship Disclosures,” International Journal of Gavilanes, José Manuel, Tessa Christina Flatten, and Malte Brettel
Advertising, 34 (4), 576–92. (2018), “Content Strategies for Digital Consumer Engagement in
Brennan, Jessica (2019), “How to Stop Wasting Your Influencer Marketing Social Networks: Why Advertising Is an Antecedent of
Budget,” Talking Influence (November 8), https://talkinginfluence.com/ Engagement,” Journal of Advertising, 47 (1), 4–23.
2019/11/08/stop-wasting-influencer-marketing-budget/. Geng, Shuang, Pianpian Yang, Yun Gao, Yingsi Tan, and
Breves, Priska Linda, Nicole Liebers, Marina Abt, and Annika Kunze Congcong Yang (2021), “The Effects of Ad Social and Personal
(2019), “The Perceived Fit Between Instagram Influencers and the Relevance on Consumer Ad Engagement on Social Media: The
Endorsed Brand: How Influencer–Brand Fit Affects Source Moderating Role of Platform Trust,” Computers in Human
Credibility and Persuasive Effectiveness,” Journal of Advertising Behavior, 122 (September), 106834.
Research, 59 (4), 440–54. Gill, Manpreet, Shrihari Sridhar, and Rajdeep Grewal (2017), “Return
Bruce, Norris I., Maren Becker, and Werner Reinartz (2020), on Engagement Initiatives: A Study of a Business-to-Business
“Communicating Brands in Television Advertising,” Journal of Mobile App,” Journal of Marketing, 81 (4), 45–66.
Marketing Research, 57 (2), 236–56. Goldenberg, Jacob, Sangman Han, Donald R. Lehmann, and Jae
Bruce, Norris I., Natasha Zhang Foutz, and Ceren Kolsarici (2012), Weon Hong (2009), “The Role of Hubs in the Adoption
“Dynamic Effectiveness of Advertising and Word of Mouth in Process,” Journal of Marketing, 73 (2), 1–13.
Sequential Distribution of New Products,” Journal of Marketing Gong, Shiyang, Juanjuan Zhang, Ping Zhao, and Xiuping Jiang (2017),
Research, 49 (4), 469–86. “Tweeting as a Marketing Tool: A Field Experiment in the TV
Campbell, Margaret C. and Amna Kirmani (2000), “Consumers’ Use Industry,” Journal of Marketing Research, 54 (6), 833–50.
of Persuasion Knowledge: The Effects of Accessibility and Grice, H.P. (1975), “Logic and Conversation,” in Syntax and
Cognitive Capacity on Perceptions of an Influence Agent,” Semantics: Speech Acts, Vol. 3. New York: Academic Press, 41–58.
Journal of Consumer Research, 27 (1), 69–83. Haenlein, Michael, Ertan Anadol, Tyler Farnsworth, Harry Hugo,
Casaló, Luis V., Carlos Flavián, and Sergio Ibáñez-Sánchez (2020), Jess Hunichen, and Diana Welte (2020), “Navigating the New
“Influencers on Instagram: Antecedents and Consequences of Opinion Era of Influencer Marketing: How to Be Successful on Instagram,
Leadership,” Journal of Business Research, 117 (September), 510–19. TikTok, & Co,” California Management Review, 63 (1), 5–25.
Castaño, Raquel, Mita Sujan, Manish Kacker, and Harish Sujan (2008), Heckman, James (1979), “Sample Selection Bias as a Specification
“Managing Consumer Uncertainty in the Adoption of New Error,” Econometrica, 47 (1), 153–61.
Products: Temporal Distance and Mental Simulation,” Journal of Hinz, Oliver, Bernd Skiera, Christian Barrot, and Jan U. Becker (2011),
Marketing Research, 45 (3), 320–36. “Seeding Strategies for Viral Marketing: An Empirical
Danaher, Peter J. and Harald J. van Heerde (2018), “Delusion in Comparison,” Journal of Marketing, 75 (6), 55–71.
Attribution: Caveats in Using Attribution for Multimedia Budget Hobbs, Bill (2019), “Understand Your Audience for Influencer
Allocation,” Journal of Marketing Research, 55 (5), 667–85. Marketing Success,” Forbes (September 16), https://www.forbes.
Datta, Hannes, Kusum L. Ailawadi, and Harald J. van Heerde (2017), com/sites/forbesbusinessdevelopmentcouncil/2019/09/16/
“How Well Does Consumer-Based Brand Equity Align with understand-your-audience-for-influencer-marketing-success/.
Sales-Based Brand Equity and Marketing-Mix Response?” Hollebeek, Linda D., Mark S. Glynn, and Roderick J. Brodie (2014),
Journal of Marketing, 81 (3), 1–20. “Consumer Brand Engagement in Social Media: Conceptualization,
22 Journal of Marketing 0(0)

Scaled Development and Validation,” Journal of Interactive Perceived Sincerity, Consumer Envy, and Product Type,” Journal
Marketing, 28 (2), 149–65. of Interactive Advertising, 20 (1), 76–91.
Hosie, Rachel (2019), “Why Brands Are Turning Away from Big Leung, Fine F., Flora F. Gu, and Robert W. Palmatier (2022), “Online
Instagram Influencers to Work with People Who Have Small Influencer Marketing,” Journal of the Academy of Marketing
Followings Instead,” Business Insider (April 9), https://www. Science, 50, 226–51.
businessinsider.com/brands-turning-to-micro-influencers-instead- Libai, Barak, Eitan Muller, and Renana Peres (2013), “Decomposing
of-instagram-stars-2019-4. the Value of Word-of-Mouth Seeding Programs: Acceleration
Hughes, Christian, Vanitha Swaminathan, and Gillian Brooks (2019), Versus Expansion,” Journal of Marketing Research, 50 (2),
“Driving Brand Engagement Through Online Social Influencers: 161–76.
An Empirical Investigation of Sponsored Blogging Campaigns,” Lind, Jo Thori and Halvor Mehlum (2010), “With or Without U? The
Journal of Marketing, 83 (5), 78–96. Appropriate Test for a U-Shaped Relationship,” Oxford Bulletin of
Humphreys, Ashlee and Rebecca Jen-Hui Wang (2018), “Automated Economics and Statistics, 72 (1), 109–18.
Text Analysis for Consumer Research,” Journal of Consumer Lou, Chen and Shupei Yuan (2019), “Influencer Marketing: How
Research, 44 (6), 1274–306. Message Value and Credibility Affect Consumer Trust of
Hyder, Shama (2019), “How to Launch a New Product or Service: Branded Content on Social Media,” Journal of Interactive
What the Latest Research Teaches Us About Successful Advertising, 19 (1), 58–73.
Launches,” Forbes (October 17), https://www.forbes.com/sites/ Lovett, Mitchelle J., Renana Peres, and Ron Shachar (2013), “On
shamahyder/2019/10/17/how-to-launch-a-new-product-or-service- Brands and Word of Mouth,” Journal of Marketing Research, 50
what-the-latest-research-teaches-us-about-successful-launches/. (4), 427–44.
Influencer Marketing Hub (2022), “The State of Influencer Malhotra, Arvind, Claudia K. Malhotra, and Alan See (2013), “How to
Marketing 2022: Benchmark Report,” (accessed March 30, Create Brand Engagement on Facebook,” MIT Slogan Management
2022), https://influencermarketinghub.com/ebooks/Influencer_ Review, 54 (2), 18–20.
Marketing_Benchmark_Report_2022.pdf. McAlister, Leigh (1982), “A Dynamic Attribute Satiation Model of
Ishihara, Masakazu and Andrew T. Ching (2019), “Dynamic Demand for Variety-Seeking Behavior,” Journal of Consumer Research, 9
New and Used Durable Goods Without Physical Depreciation: The (2), 141–50.
Case of Japanese Video Games,” Marketing Science, 38 (3), 392–416. McQuarrie, Edward F., Jessica Miller, and Barbara J. Phillips (2013),
Jaccard, James and Robert Turrisi (2003), Interaction Effects in Multiple “The Megaphone Effect: Taste and Audience in Fashion
Regression, 2nd ed. Thousand Oaks, CA: SAGE Publications. Blogging,” Journal of Consumer Research, 40 (1), 136–58.
Ki, Chung-Wha Chloe and Youn-Kyung Kim (2019), “The Mechanism by Min, Sungwook, Namwoon Kim, and Ge Zhan (2017), “The Impact of
Which Social Media Influencers Persuade Consumers: The Role of Market Size on New Market Entry: A Contingency Approach,”
Consumers’ Desire to Mimic,” Psychology & Marketing, 36 (10), European Journal of Marketing, 51 (1), 2–22.
905–22. Moldovan, Sarit, Jacob Goldenberg, and Amitava Chattopadhyay
Krishnan, Trichy V. and Dipak C. Jain (2006), “Optimal Dynamic (2011), “The Different Roles of Product Originality and
Advertising Policy for New Products,” Management Science, 52 Usefulness in Generating Word-of-Mouth,” International Journal
(12), 1957–69. of Research in Marketing, 28 (2), 109–19.
Kupfer, Ann-Kristin, Nora Pähler vor der Holte, Raoul V. Kübler, and Nan, Xiaoli and Ronald J. Faber (2004), “Advertising Theory:
Thorsten Hennig-Thurau (2018), “The Role of the Partner Brand’s Reconceptualizing the Building Blocks,” Marketing Theory, 4 (1/2),
Social Media Power in Brand Alliances,” Journal of Marketing, 82 7–30.
(3), 25–44. Nelson, Jacob L. and James G. Webster (2016), “Audience Currencies
Labrecque, Lauren, Jonas Vor Dem Esche, Charla Mathwick, Thomas in the Age of Big Data,” International Journal on Media
P. Novak, and Charles F. Hofacker (2013), “Consumer Power: Management, 18 (1), 9–24.
Evolution in the Digital Age,” Journal of Interactive Marketing, Neumann, Nico, Catherine E. Tucker, and Timothy Whitfield
27 (4), 257–69. (2019), “Frontiers: How Effective is Third-Party Consumer
Lanz, Andreas, Jacob Goldenberg, Daniel Shapira, and Florian Stahl Profiling? Evidence from Field Studies,” Marketing Science,
(2019), “Climb or Jump: Status-Based Seeding in User-Generated 38 (6), 918–26.
Content Networks,” Journal of Marketing Research, 56 (3), 361–78. Nunes, Joseph C., Andrea Ordanini, and Gaia Giambastiani (2021),
Lasswell, Harold D. (1948), “The Structure and Function of Communication “The Concept of Authenticity: What It Means to Consumers,”
in Society,” The Communication of Ideas, 37 (1), 136–39. Journal of Marketing, 85 (4), 1–20.
Lee, Dokyun, Kartik Hosanagar, and Harikesh S. Nair (2018), “Advertising Ordenes, Francisco Villarroel, Dhruv Grewal, Stephen Ludwig, Ko De
Content and Consumer Engagement on Social Media: Evidence from Ruyter, Dominik Mahr, and Martin Wetzels (2018), “Cutting
Facebook,” Management Science, 64 (11), 5105–31. Through Content Clutter: How Speech and Image Acts Drive
Lee, Jeffrey K. and Enric Junqué de Fortuny (2022), Consumer Sharing of Social Media Brand Messages,” Journal of
“Influencer-Generated Reference Groups,” Journal of Consumer Consumer Research, 45 (5), 988–1012.
Research, 49 (1), 25–45. Pansari, Anita and V. Kumar (2017), “Customer Engagement: The
Lee, Jung Ah and Matthew S. Eastin (2020), “I Like What She’s Construct, Antecedents, and Consequences,” Journal of the
#Endorsing: The Impact of Female Social Media Influencers’ Academy of Marketing Science, 45 (3), 294–311.
Leung et al. 23

Peers, Yuri, Harald J. van Heerde, and Marnik G. Dekimpe (2017), Companies’ Twitter Communications,” Industrial Marketing
“Marketing Budget Allocation Across Countries: The Role of Management, 43 (5), 873–81.
International Business Cycles,” Marketing Science, 36 (5), 792–809. Teixeira, Thales S., Michel Wedel, and Rik Pieters (2010),
Pei, Amy and Dina Mayzlin (2021), “Influencing Social Media “Moment-to-Moment Optimal Branding in TV Commercials:
Influencers Through Affiliation,” Marketing Science, 41 (3), 593–615. Preventing Avoidance by Pulsing,” Marketing Science, 29 (5), 783–804.
Pennebaker, James W., Ryan L. Boyd, Kayla Jordan, and Kate Blackburn Tellis, Gerard J., Deborah J. MacInnis, Seshadri Tirunillai, and
(2015), “The Development and Psychometric Properties of Yanwei Zhang (2019), “What Drives Virality (Sharing) of Online
LIWC2015,” working paper, University of Texas at Austin. Digital Content? The Critical Role of Information, Emotion, and
Petty, Richard E. and John T. Cacioppo (1986), “The Elaboration Brand Prominence,” Journal of Marketing, 83 (4), 1–20.
Likelihood Model of Persuasion,” Advances in Experimental Tirunillai, Seshadri and Gerard J. Tellis (2012), “Does Chatter Really
Social Psychology, 19 (C), 123–205. Matter? Dynamics of User-Generated Content and Stock
Rooderkerk, Robert P. and Koen H. Pauwels (2016), “No Comment?! Performance,” Marketing Science, 31 (2), 198–215.
The Drivers of Reactions to Online Posts in Professional Groups,” Uribe, Rodrigo, Cristian Buzeta, and Milendka Velásquez (2016),
Journal of Interactive Marketing, 35 (1), 1–15. “Sidedness, Commercial Intent, and Expertise in Blog
Self, Charles C. (2009), “Credibility,” in An Integrated Approach to Advertising,” Journal of Business Research, 69 (10), 4403–10.
Communication Theory and Research. New York: Routledge, 435–56. Valsesia, Francesca, Davide Proserpio, and Joseph C. Nunes (2020),
Sethuraman, Raj, Gerard J. Tellis, and Richard A. Briesch (2011), “The Positive Effect of Not Following Others on Social Media,”
“How Well Does Advertising Work? Generalizations from Journal of Marketing Research, 57 (6), 1152–68.
Meta-Analysis of Brand Advertising Elasticities,” Journal of Van Heerde, Harald J., Maarten J. Gijsenberg, Marnik G. Dekimpe,
Marketing Research, 48 (3), 457–71. and Jan-Benedict E.M. Steenkamp (2013), “Price and Advertising
Shannon, Claude E. and Warren Weaver (1949), The Mathematical Effectiveness over the Business Cycle,” Journal of Marketing
Theory of Communication. Urbana: University of Illinois Press. Research, 50 (2), 177–93.
Soboleva, Alena, Suzan Burton, Girijasankar Mallik, and Varnali, Kaan (2021), “Online Behavioral Advertising: An Integrative
Aila Khan (2017), “Retweet for a Chance to …: An Analysis Review,” Journal of Marketing Communications, 27 (1), 93–114.
of What Triggers Consumers to Engage in Seeded eWOM on Venkatraman, Vinod, Angelika Dimoka, Paul A. Pavlou, Khoi Vo,
Twitter,” Journal of Marketing Management, 33 (13/14), William Hampton, Bryan Bollinger, et al. (2015), “Predicting
1120–48. Advertising Success Beyond Traditional Measures: New Insights
Sridhar, Shrihari, Frank Germann, Charles Kang, and Rajdeep Grewal from Neurophysiological Methods and Market Response
(2016), “Relating Online, Regional, and National Advertising to Modeling,” Journal of Marketing Research, 52 (4), 436–52.
Firm Value,” Journal of Marketing, 80 (4), 39–55. Walker, Lorna, Paul R. Baines, Radu Dimitriu, and Emma
Stephen, Andrew T., Yaniv Dover, Lev Muchnik, and K. Macdonald (2017), “Antecedents of Retweeting in a (Political)
Jacob Goldenberg (2017), “Pump It Out! The Effect of Marketing Context,” Psychology & Marketing, 34 (3), 275–93.
Transmitter Activity on Content Propagation in Social Media,” Wielgos, Dominik M., Christian Homburg, and Christina Kuehnl
working paper, Saïd Business School, University of Oxford. (2021), “Digital Business Capability: Its Impact on Firm and
Stubb, Carolina (2018), “Story Versus Info: Tracking Blog Readers’ Customer Performance,” Journal of the Academy of Marketing
Online Viewing Time of Sponsored Blog Posts Based on Science, 49, 762–89.
Content-Specific Elements,” Computers in Human Behavior, 82 Wilson, Elizabeth J. and Daniel L. Sherrell (1993), “Source Effects in
(May), 54–62. Communication and Persuasion Research: A Meta-Analysis of
Suh, Bongwon, Lichan Hong, Peter Pirolli, and Ed H. Chi (2010), Effect Size,” Journal of the Academy of Marketing Science, 21
“Want to Be Retweeted? Large Scale Analytics on Factors (2), 101–12.
Impacting Retweet in Twitter Network,” 2010 IEEE Second Yuan, Shupei and Chen Lou (2020), “How Social Media Influencers
International Conference on Social Computing, 177–84. Foster Relationships with Followers: The Roles of Source
Swani, Kunal, Brian P. Brown, and George R. Milne (2014), “Should Credibility and Fairness in Parasocial Relationship and Product
Tweets Differ for B2B and B2C? An Analysis of Fortune 500 Interest,” Journal of Interactive Advertising, 20 (2), 133–47.

View publication stats