d-angelo-valsesia-2022-you-should-try-these-together-combinatory-recommendations-signal-expertise-and-improve-product

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Article

Journal of Marketing Research


2023, Vol. 60(1) 155-169
You Should Try These Together: Combinatory © American Marketing Association 2022
Article reuse guidelines:

Recommendations Signal Expertise sagepub.com/journals-permissions


DOI: 10.1177/00222437221111344
journals.sagepub.com/home/mrj
and Improve Product Attitudes

Jennifer K. D’Angelo and Francesca Valsesia

Abstract
This work introduces a novel cue that consumption advisers, like stylists and interior designers, can use to signal expertise:
combinatory recommendations. In a combinatory recommendation, a person offers an opinion about compatibility among mul-
tiple products intended for joint usage. Across nine studies conducted in the lab and field, the authors find that offering a com-
binatory recommendation signals greater expertise (Study 1a, Study 2a) and, specifically, greater depth of knowledge (Study 1b),
compared with other types of recommendations involving the same number of products. This effect does not depend on the
helpfulness of the adviser (Study 2b) but is qualified by features of the recommendation itself (Study 3a) as well as the type
of combination recommended (Study 3b). Importantly, the authors find this effect to have important downstream consequences,
as the increased perceptions of expertise that follow a combinatory recommendation improve consumers’ attitudes both toward
products included in the recommendation and toward subsequent recommendations made by the adviser (Study 4, Study 5). The
real-world persuasive value of combinatory recommendations is also tested in a field study (Study 6) that explores the effect of
combinatory recommendations on click-through rates of Instagram advertisements.

Keywords
signaling, expertise, combinations, complements, recommendations
Online supplement: https://doi.org/10.1177/00222437221111344

In recent years, brands have made significant investments in combinatory recommendation, a person offers an opinion
enlisting consumption advisers to recommend their products about compatibility among multiple products considered for
to consumers (Gorin 2018; Howland 2018). These advisers joint usage (i.e., whether or not these products go well together)
exist in a breadth of product domains, ranging from stylists within a single consumption occasion. For instance, a stylist
who facilitate consumers’ shopping experiences (Bobb 2019; might discuss how a certain tie pairs well with a dress shirt to
Davenport et al. 2020) to social media influencers who recom- curate an outfit; an interior designer might suggest hanging dif-
mend products to their followers (Libai, Muller, and Peres ferent pieces of artwork together to create a gallery wall; a
2013; Valsesia, Proserpio, and Nunes 2020). Even at the beauty influencer might suggest combining multiple skin-care
height of the COVID-19 pandemic, retailers doubled down on products to form a skin-care routine. Combinatory recommen-
the usage of these advisers. For example, retailers like dations may also provide opinions about what does not go
Neiman Marcus and John Lewis launched services whereby well together. For example, a chef might warn against pairing
consumers could interact with online stylists and interior sparkling wine with spicy fish.
designers (Coker 2020; Wilson 2020). Further, in 2020, most This research focuses on situations where an adviser offers a
U.S. brands increased their influencer spending despite consumer an unsolicited combinatory recommendation. We
cutting a variety of other marketing expenses (Glenday 2020). propose that such combinatory recommendations reflect an
Although consumption advisers come in many forms, they ability to process how multiple products will interact with one
are related in their motivation to signal expertise to gain con- another, which consumers perceive as a demonstration of the
sumers’ trust and increase their persuasiveness (Alba and
Hutchinson 1987), thereby differentiating their services in the
Jennifer K. D’Angelo is Assistant Professor of Marketing, Neeley School of Business,
ever-growing marketplace of online and offline advising ser- Texas Christian University, USA (email: jennifer.dangelo@tcu.edu). Francesca Valsesia
vices. In this work, we introduce a novel cue advisers can (corresponding author) is Assistant Professor of Marketing, Foster School of
use to signal expertise: combinatory recommendations. In a Business, University of Washington, USA (email: valsesia@uw.edu).
156 Journal of Marketing Research 60(1)

adviser’s depth of knowledge in a product category. Thus, this recommendations as a tactic advisers can implement to signal
adviser signals expertise in the category. Moreover, their rec- their expertise to consumers and differentiate themselves in
ommendation may garner more favorable attitudes from con- an increasingly crowded marketplace.
sumers, consistent with findings that experts are more
influential and persuasive (Alba and Hutchinson 1987;
Gershoff, Broniarczyk, and West 2001). Theoretical Framework
The literature shows that consumers rely on a variety of cues
Prior literature highlights how consumers can enjoy products
to infer others’ expertise, including the individual’s appearance more when consuming them in combination, especially if
(Bellezza, Gino, and Keinan 2014), choices (Sela et al. 2019), these products appear to be made for each other, provide a
and demeanor (Price and Stone 2004; Sniezek and Buckley
novel consumption experience, or appease intricate preferences
1995), as well as the consistency between the individual’s (Hildebrand, Häubl, and Herrmann 2014; Rahinel and Redden
opinion and other external sources of information (Palmeira 2013; Spence 2020a). With a marketplace surge in customiza-
2020). These cues can be used strategically to signal expertise.
tion opportunities, consumers now have many opportunities
Our research extends the literature on signaling and persuasion to combine products for themselves (D’Angelo, Diehl, and
by identifying combinatory recommendations as a novel cue Cavanaugh 2019). Yet, products typically have and continue
that signals expertise and increases persuasiveness. In doing
to be combined by the marketer (Karataş and Gürhan-Canli
so, we also contribute to the literature on consumer knowledge 2020; Tellis 1986), alluding to the notion that lay consumers
and expertise. Prior work generally categorizes depth of knowl- may lack an understanding of compatibility (i.e., which prod-
edge as understanding the details of a product’s attributes in
ucts go well together), at least in some product categories.
isolation (Alba and Hutchinson 1987). We show that displaying This lack of understanding may stem from the cognitive
an understanding of interactions across the attributes of differ- demand consumers encounter when simultaneously evaluating
ent products also contributes to the perception that an individual
multiple products (Aribarg and Foutz 2009) and the difficulties
has depth of knowledge in a product category, resulting in in predicting the outcome of product combinations (Deng, Hui,
greater perceived expertise. and Wesley Hutchinson 2010; Petre, Sharp, and Johnson 2006).
Combinatory recommendations relate to two important
Many things could happen when products interact. Indeed, the
constructs: curation and bundling. The nascent marketing lit- outcome of a product’s attributes can differ depending on what
erature on curation uses the term to refer to the selection, it is combined with, as these attributes could be enhanced,
organization, or presentation of a combination of objects
emphasized, or suppressed (Smith and Redden 2020). The
(Babin and Hulland 2019). Thus, curation does not necessar- outcome, in some cases, may even be nonmonotonic, changing
ily imply the presence of an explicit recommendation nor the over the course of the consumption occasion. For example,
inclusion of items intended for joint usage. In this sense, a
when combining salt and tonic water, the mixture first tastes
combinatory recommendation could be seen as a particular sweet but becomes noticeably saltier over time (Spence
type of curation. Little is known about consumer responses 2020b). We thus argue that consumers perceive an individual
to those in charge of curation and to the curated selection
who seemingly demonstrates an understanding of product inter-
of products. We suggest that exhibiting an ability to curate actions via a combinatory recommendation as having expertise
products for joint usage can be an effective signal of in a product category.
expertise.
Bundling, in contrast, refers to “marketing two or more prod-
ucts and/or services in a single package for a special price” Dimensions of Expertise Signaled by Combinatory
(Guiltinan 1987, p. 74). Our research encompasses a broader
lens, including situations where products are not packaged Recommendations
together by marketers nor sold at a special price. Nonetheless, Experts differ from novices in both the amount and organization
a bundle of complementary products can be characterized as of their domain knowledge, and domain knowledge is usually
one type of combinatory recommendation. For this reason, measured to assess expertise (for a discussion, see Mitchell
our work also contributes to the bundling literature. Prior and Dacin [1996]). Two knowledge dimensions contribute to
work largely focuses on consumers’ evaluations of the items expertise: breadth and depth of knowledge. Breadth refers to
included in a bundle (Karataş and Gürhan-Canli 2020; the diverse knowledge of various options in a category,
Rahinel and Redden 2013) rather than on consumers’ percep- whereas depth refers to a focused understanding of each indi-
tions of the bundle curator. Our work is consistent with findings vidual option (Clarkson, Janiszewski, and Cinelli 2013;
that consumers tend to favorably evaluate complementary Mannucci and Yong 2018; Yang, Jin, and Sheng 2017).
product bundles over other types of bundles (Popkowski When evaluating someone’s expertise, consumers seek cues
Leszczyc and Haubl 2010; Zhao and Xia 2020) and offers a indicating that the individual has a breadth and/or a depth of
contributory explanation driving these favorable evaluations: knowledge in the category (Sela et al. 2019). We posit that a
perceived expertise of the bundle curator. combinatory recommendation signals expertise to consumers
In addition to these theoretical contributions, we make a sub- because it is perceived as a demonstration of depth of
stantive contribution by introducing the use of combinatory knowledge.
D’Angelo and Valsesia 157

Prior work focuses on depth of knowledge of individual Factors That Influence the Signaling Effectiveness
product attributes (e.g., Clarkson, Janiszewski, and Cinelli of Combinatory Recommendations
2013) and discusses how the ability to analyze attributes is a
manifestation of knowledge depth (Alba and Hutchinson Joint usage could manifest through various means, such as a
1987). We build on this work and argue that the ability to blending, fusion, or pairing of multiple, discrete products
analyze interactions of attributes among multiple products is (Spence 2020a). Our theoretical account can apply to these
a unique form of knowledge depth. This notion is supported various forms of joint usage, which we assess empirically.
by work in psychology suggesting that some further factor Nonetheless, the extent to which an adviser signals expertise
must be included in cognition when one engages in integrative when providing a combinatory recommendation should
thinking about how elements interact (Kallio 2011). For depend on the extent to which the recommendation highlights
example, when evaluating the compatibility of paprika and the interaction among product attributes. Cue utilization
cumin for joint usage, one considers not only the individual theory suggests that the extent to which a signal is used by con-
spices’ attributes but also how these attributes will interact sumers to form an evaluation varies as a function of its ambigu-
(e.g., determining whether the combination will result in an ity (Slovic and Lichtenstein 1971). It is therefore reasonable to
overpowering amount of spiciness). Thus, using combinatory expect that an explicit combinatory recommendation, in which
recommendations, where one seemingly demonstrates the an adviser speaks about the interaction, will more clearly dem-
ability to understand outcomes of product interactions, may onstrate depth of knowledge compared with an implicit combi-
reflect depth of knowledge and thereby signal expertise. natory recommendation.
Formally:
H2: The same combinatory recommendation leads to
greater expertise perceptions if it explicitly addresses
H1a: Keeping the number of products constant, providing product attribute interactions.
recommendations about the joint (vs. nonjoint) usage of
these products signals greater expertise. Moreover, not all combinatory recommendations imply the
same level of interaction between product attributes (Spence
H1b: Keeping the number of products constant, providing 2020a). Think, for instance, of recommending two products
recommendations about the joint (vs. nonjoint) usage of for a meal that are to be consumed simultaneously (two food
these products reflects greater depth (but not breadth) of items served on the same plate) versus sequentially (two food
knowledge. items served on different plates, one after the other).
Although both could be framed as a combinatory recommenda-
tion, the temporal distance between individual consumption
Instead of recommending multiple products for joint usage, acts (eating a food item) within the single consumption occa-
an adviser might make recommendations that do not sion (the meal) is longer in the latter case. Similarly, the phys-
discuss joint usage (i.e., using products during separate ical proximity of the products recommended for joint usage
consumption occasions or as substitutes). According to (pairing a tie with a shirt vs. pairing a tie with socks) affects
prior literature, which typically dichotomizes products the spatial distance between products. We reason that the
that share some relation as either complements or substi- greater the distance between products or consumption acts,
tutes (Diehl, Van Herpen, and Lamberton 2015; Huh, the lower the expected interaction between product attributes.
Vosgerau, and Morewedge 2016; Karataş and For this reason:
Gürhan-Canli 2020; Shocker, Bayus, and Kim 2004), a
natural comparison to combinatory recommendations H3: The effect of a combinatory recommendation on per-
would be substitute recommendations. By our theoretical ceptions of expertise is reduced the greater the distance
account, when evaluating two products intended for use (either temporal or spatial) between individual consump-
in separate consumption occasions or when two products’ tion acts.
substitutability is assessed, an understanding of product
interactions is not needed, thus signaling less expertise The Persuasive Effect of Combinatory Recommendations
than combinatory recommendations signal. Perceived expertise is a peripheral cue known to affect persua-
Note that expertise cues are based on lay beliefs of how sive outcomes (Petty and Cacioppo 1986). Therefore, in line
an expert should behave but may not necessarily reflect with the literature on expertise (Alba and Hutchinson 1987;
real expertise (Packard and Wooten 2013). Therefore, Gershoff, Broniarczyk, and West 2001), we predict that when
although compatibility can be prescribed by objective orga- an adviser signals expertise by using a combinatory recommen-
nizing principles (e.g., extent of harmony among the prod- dation, they become more persuasive. As such, that adviser
ucts; Spence 2020a), we argue that merely recommending should garner more favorable consumer attitudes toward their
products for joint usage, even if the recommendation is recommendations. This includes products mentioned in the
not based on objective principles, may signal expertise to combinatory recommendation as well as subsequent, unrelated
consumers. product recommendations.
158 Journal of Marketing Research 60(1)

H4: An individual who uses a combinatory recommenda- a stronger signal of depth of knowledge. Finding that a combi-
tion (vs. not) is more persuasive because a combinatory natory recommendation has a stronger effect on signaling
recommendation signals expertise. expertise would support our theorizing. Note that although we
include the same quantity of information in the substitute and
This research tests our predictions across nine studies con- combinatory recommendations (see the Web Appendix for a
ducted in the lab and field. We generalize our findings pilot study), these recommendations should be perceived as
across a breadth of product categories, including categories qualitatively different, as a combinatory recommendation
where the products can be experienced (e.g., wall art) and should signal greater expertise.
cannot be experienced (e.g., spices) as the recommendation
is being made. We also generalize our findings across
several types of advisers (e.g., subscription box curator, Method
store associate, interior designer, influencer). Study 1a
A total of 300 participants from Amazon Mechanical Turk
finds that recommending two products that go well together
(MTurk) completed the study for a nominal payment.1
(i.e., a combinatory recommendation) signals greater exper-
Participants on MTurk are known to misrepresent their char-
tise than recommending two products that could be used as
acteristics to meet study eligibility (Chandler and Paolacci
substitutes or recommending a single product (H1a). Study
2017). Therefore, MTurk participants were only allowed
1b digs deeper into the type of expertise signaled and finds
to complete the study if they successfully passed a U.S. cul-
that a combinatory recommendation signals greater depth
tural check question and an attention check question. Full
(but not breadth of knowledge, H1b). Studies 2a and 2b
details of these questions are reported in the Web
address potential alternative accounts for the signaling
Appendix. We also allowed only one response per MTurk
value of combinatory recommendations. Study 2a finds that
ID. If participants attempted the cultural check or the atten-
combinatory recommendations that (1) suggest which prod-
tion check questions multiple times, or entered the study
ucts go well together or (2) warn of which products do not
twice for other reasons, only their first answer, even if
go well together are equally effective in signaling expertise,
incomplete, was considered valid. Subsequent responses
thus showing that the effect of combinatory recommenda-
were excluded from the analysis, resulting in a final
tions on perceived expertise does not depend on the utility
sample of 286 responses (Mage = 39 years, 44% female).
consumers can extract from the purchase. Study 2b shows
The same exclusion criteria were used across all our
that the effect does not depend on whether the recommender
studies conducted online.
is addressing a specific consumer need. Studies 3a and 3b
Participants were randomly assigned to one of three rec-
address the moderating role of explicitness of the combina-
ommendation conditions (combinatory, substitute, single)
tory recommendation (H2) and of the distance between rec-
using a between-subjects design. Participants imagined
ommended consumption acts (H3).
signing up for PantryBox, a monthly subscription box
The next studies explore the downstream consequences
company that delivers a selection of spices and other
of the effect of combinatory recommendations. Study 4
cooking ingredients. The subscription box included a note
finds that a combinatory recommendation improves con-
from Chef Taylor, the curator of the subscription box. The
sumers’ attitudes toward a product from that recommenda-
chef’s note highlighted some of the products included in
tion and provides mediation evidence of perceived
the box. The number of products participants were told
expertise of the adviser as the underlying driver of this
were included in the box did not differ across conditions,
effect (H 4). Study 5 finds that, following an adviser’s com-
although the number of products highlighted in the note
binatory recommendations, consumers’ attitudes toward
did differ (see the Web Appendix for full descriptions of
subsequent, unrelated product recommendations also
stimuli for all studies).
improve. Finally, Study 6 uses an Instagram ad field study
Specifically, in the combinatory condition the chef’s note
to demonstrate the real-world persuasive value of combina-
described two spices, stating, “You have to try our cardamom.
tory recommendations.
It has an addicting, tangy and herby freshness to it. Its flavor
profile pairs beautifully with the sweetness of fenugreek
seeds. Together, they make a balanced combination. Using
Study 1a: Combinatory Recommendations these two complimentary spices together will add great
Signal Expertise flavors to your dish.”
The substitute condition’s note also described two spices.
Study 1a tests our core prediction that an adviser who provides
The note read, “You have to try our cardamom. It has an addict-
a combinatory recommendation signals expertise. We compare
ing, tangy and herby freshness to it. If you are not looking for
a combinatory recommendation with a substitute recommenda-
this flavor profile, try the sweetness of fenugreek seeds. They
tion involving the same number of products (i.e., two). We posit
that determining compatibility (compared with substitutability)
requires a more complex processing of interactions among 1
For all studies, we aimed to recruit at least 100 participants per cell. All con-
products, and thus a combinatory recommendation should be ditions are reported, and studies received Institutional Review Board approval.
D’Angelo and Valsesia 159

are both delicious spices. Using either one of these spices will Study 1b: Combinatory Recommendations
add great flavors to your dish.” Affect Depth, but Not Breadth, of Knowledge
Finally, in the single condition, the chef only described the
Perceptions
flavor profile of one spice and how that spice would add great
flavor to a dish. The one spice described in this condition was We compare a combinatory recommendation with a substitute
counterbalanced as either cardamom or fenugreek seeds. We recommendation involving the same number of products and
included this condition as a baseline for this first study. examine the extent to which each cues breadth and depth of
Subsequent studies focus exclusively on comparing situations knowledge. We expect combinatory and substitute recommen-
where the number of products recommended is the same, in dations to reflect similar levels of breadth of knowledge.
line with our theorizing. However, providing a combinatory recommendation suggests
To measure to what extent Chef Taylor’s recommendation that a person can process the outcome of interactions among
signaled category expertise, participants rated the expertise of products, which we suggest is a stronger cue of depth of
the chef using a three-item measure (person seems like an knowledge.
expert on food, person seems knowledgeable about food,
person knows what they are talking about) on a nine-point
scale (1 = “Not at all,” and 9 = “Very much”). The items were
Method
averaged into one composite measure of perceived expertise Participants from MTurk (N = 199) completed the study for a
(α = .96). For all studies, all dependent measures taken are nominal payment. One response from a duplicate ID was
reported (see the Web Appendix for full wording of all mea- excluded, leaving a sample of 198 responses (Mage = 38
sures taken). years, 57% female).
Participants were randomly assigned to one of two recom-
mendation conditions (combinatory vs. substitute) using a
Results between-subjects design. They imagined that while they were
In the single condition, there was no significant difference in browsing wall art at a furniture store, the store’s interior
perceived expertise, regardless of whether the chef described design adviser provided them with an unsolicited recommenda-
cardamom or fenugreek seeds (F < 1); thus, responses were col- tion. In both conditions, the adviser discussed two pieces of
lapsed into one condition. wall art, one depicting the Isle of Skye and one depicting the
A one-way analysis of variance (ANOVA) revealed a signif- Botallack Tin Mine. Both pieces of art were shown to the
icant effect of recommendation on perceived expertise (F(2, participant.
283) = 26.62, p < .001, ω2p = .152). Compared with the single In the combinatory recommendation condition, the
condition (M = 5.72), perceptions of expertise were greater in adviser noted how the Botallack Tin Mine art would be a
both the combinatory condition (M = 7.37; F(1, 283) = 53.23, good complement to the Isle of Skye art. In contrast, in the
p < .001, ω2p = .155) and the substitute condition (M = 6.57, substitute condition, the adviser noted how the Botallack
F(1, 283) = 13.47, p < .001, ω2p = .042). Importantly, as pre- Tin Mine art would be a good alternative to the Isle of
dicted, the combinatory recommendation also signaled signifi- Skye art. After reading the recommendation, participants
cantly more expertise than the substitute recommendation rated the depth of knowledge possessed by the interior
(F(1, 283) = 12.45, p < .001, ω2p = .039). designer using a three-item measure (person understands a
lot about these pieces of wall art, person seems to know
many details about these pieces of wall art, person has a deep
Discussion understanding of these pieces of wall art) on a seven-point
Study 1a shows that providing a combinatory recommendation scale (1 = “Strongly disagree,” and 7 = “Strongly agree”).
signals greater expertise than recommending a single product, Participants then rated the breadth of knowledge possessed by
perhaps by demonstrating a greater breadth of knowledge of the interior designer using a three-item measure (person is
products in the category. More pertinent to our theorizing, familiar with many wall art options, person seems to know at
and in support of H1b, we find that providing a combinatory rec- least a little bit about lots of different types of wall art, person
ommendation signals greater expertise than recommending the has some knowledge about a great number of wall art pieces)
same number of products as substitutes. Given that the number on a seven-point scale (1 = “Strongly disagree,” and 7 =
of items recommended in these recommendations is the same, “Strongly agree”). We validated these depth and breadth
the breadth of knowledge displayed is presumably similar. items as separate factors in a pilot study reported in the Web
We suggest that the difference in expertise perceptions may Appendix.
come from a combinatory recommendation displaying one’s
understanding of the outcome of interactions among products
and thus demonstrating depth of knowledge. In the next Results
study, we examine breadth and depth of knowledge to better The Fornell and Larcker (1981) criterion for discriminant
understand how combinatory recommendations affect these validity requires the average variance extracted (AVE) of
components of expertise. both constructs to be greater than the squared correlation
160 Journal of Marketing Research 60(1)

between the two constructs. In this case, the depth AVE is combinatory recommendation differs in signaling expertise
.86, the breadth AVE is .88, and the squared correlation compared with a negative combinatory recommendation,
between the two variables is .38, meeting this criterion where products are described as incompatible. Both types of
and suggesting discriminant validity between breadth and combinatory recommendations could demonstrate depth of
depth of knowledge. Thus, the three depth-related items knowledge (i.e., one is able to determine whether or not the
were averaged into one composite measure of depth of interactions among the products would lead to positive con-
knowledge (α = .94) and the three breadth-related items sumption outcomes). Thus, we predict that both types of com-
were averaged into one composite measure of breadth of binatory recommendations can signal greater expertise
knowledge (α = .95). compared with a substitute recommendation. If, however,
Supporting H1b, a one-way ANOVA revealed that perceived expertise is signaled because of one’s ability to create superad-
depth of knowledge was greater in the combinatory recommen- ditive utility, one should signal expertise with a positive combi-
dation condition (M = 5.19), compared with the substitute rec- natory recommendation but not with a negative combinatory
ommendation condition (M = 4.71, F(1, 196) = 5.52, p = recommendation.
.020, ω2p = .022). Importantly, breadth of knowledge did not
differ between the combinatory (M = 4.43) and substitute
conditions (M = 4.58; F < 1). Method
Undergraduate students from two universities (N = 337, Mage =
20 years, 47% female) completed an online study within a
Discussion one-week period. Sample size was determined by the number
Study 1b demonstrates that the depth, but not breadth, compo- of students who signed up to take the study for course credit.
nent of expertise is heightened when comparing a combinatory There were no significant differences across the two universi-
recommendation with a substitute recommendation. ties; therefore, results were collapsed across universities.
Yet, there may be some alternative accounts for why a com- Participants were randomly assigned to one of three recom-
binatory recommendation signals greater expertise. For one, the mendation conditions (positive combinatory, negative combi-
bundling literature suggests that complementary items have natory, substitute) using a between-subjects design. As in
superadditive utilities, whereas substitute items do not. That Study 1a, participants read Chef Taylor’s note describing two
is, the utility that consumers extract from a bundle with comple- spices. Across conditions, the note’s description of the individ-
mentary items is greater than the sum of utilities provided by ual characteristics of cardamom and fenugreek seeds was held
each item in isolation (Guiltinan 1987; Karataş and constant. However, in the positive combinatory condition, the
Gürhan-Canli 2020). Thus, it is possible that when one provides note also described how the characteristics of the two spices
a combinatory recommendation, one may be demonstrating an paired beautifully and recommended using the spices together
ability to create superadditive utility of a purchase. One’s ability in a dish. In the negative combinatory condition, the note
to create superadditive utility, in turn, may signal expertise. instead described how the characteristics of the two spices did
Another alternative explanation for the results of Study 1a not pair well together and recommended not using these
may be that, while it was not explicitly mentioned, respon- spices together in a dish. In the substitute condition, the note
dents might have believed their goal when getting a subscrip- described how the characteristics of the two spices were
tion box was to purchase multiple products for joint usage, and equally delicious and recommended using either one in a
a combinatory recommendation helps achieve that goal more dish. Participants completed the same perceived expertise
so than a substitute recommendation. If so, respondents in measure used in Study 1a (α = .94).
that study may have perceived the adviser as being more
helpful, which, in turn, drove perceptions of expertise. Results
Although it is possible that, in some situations, these elements
would contribute to combinatory recommendations’ ability to A one-way ANOVA revealed a significant effect of recommenda-
signal expertise, our next two studies show that our findings tion on perceived expertise (F(2, 334) = 5.01, p = .007, ω2p = .023).
from Study 1a hold in situations where combinatory recom- A substitute recommendation (M = 6.25) signaled significantly
mendations do not provide superadditive utility and are not less expertise than both a positive combinatory (M = 6.77;
helpful to the consumer. F(1, 334) = 7.05, p = .008, ω2p = .018) and negative combina-
tory (M = 6.81; F(1, 334) = 8.02, p = .005, ω2p = .020) recom-
mendation. Importantly, positive and negative combinatory
Study 2a: Both Positive and Negative recommendations signaled similar levels of expertise (F < 1).
Combinatory Recommendations Signal
Expertise Discussion
In the previous studies, we focused on a positive combinatory Study 2a’s findings suggest that the display of one’s ability to
recommendation, where the products recommended were process both what is and what is not compatible can equally
described as compatible. Study 2a tests whether a positive signal expertise more so than a substitute recommendation.
D’Angelo and Valsesia 161

Thus, even when one’s combinatory recommendation does not Results and Discussion
demonstrate an ability to create superadditive utility, as was the
Consistent with our previous studies, the interior designer
case with the negative combinatory recommendation, one can
signaled significantly more expertise when providing a com-
still use the recommendation to signal expertise.
binatory recommendation (M = 6.69) than when providing
For the remaining studies, we focus on positive combinatory
a substitute recommendation (M = 5.59; F(1, 260) = 30.17,
recommendations, as marketers may be inclined to discuss the
p < .001, ω2p = .100). This was in spite of the interior designer
positive aspects of their products and encourage, rather than dis-
being perceived as significantly less helpful in the combinatory
suade, consumers from purchasing multiple products.
recommendation condition (M = 6.31) than in the substitute
condition (M = 7.32; F(1, 260) = 21.77, p < .001, ω2p = .073).
Study 2b: Combinatory Recommendations Thus, even in situations where an adviser is less helpful in
achieving the consumer’s goal, the use of combinatory recom-
Signal Greater Expertise Even When They
mendations can signal greater expertise.
Are Less Helpful Than Substitute So far, we focused on comparing combinatory recommenda-
Recommendations tions with substitute recommendations. However, to further test
Study 2b tests whether expertise is contingent on the help- our theorizing and demonstrate boundary conditions of combi-
fulness of the recommendation. When providing a combina- natory recommendations, the next two studies examine other
tory recommendation, we expect that an adviser should be comparison conditions beyond substitutes. First, to test H2,
able to demonstrate their depth of knowledge of the prod- we focus specifically on combinatory recommendations and
ucts in the recommendation, even if the information pro- vary how the recommendation is formulated.
vided is not particularly helpful to the consumer. Thus, we
predict that a combinatory recommendation can signal
greater expertise compared with a substitute recommenda- Study 3a: Explicit Combinatory
tion independent of how helpful the recommendation is to Recommendations Signal Greater Expertise
the consumer. Than Implicit Combinatory
Recommendations
Method Study 3a examines what happens when an adviser makes (or
does not make) their knowledge about interactions explicitly
Undergraduate students (N = 262; Mage = 20 years, 47%
known. In other words, the combination of products recom-
female) completed a lab study within a two-week period.
mended remains the same, but the emphasis put on the
Sample size was determined by the number of students who
product interactions varies. We expect that emphasizing
signed up to take the study for course credit. Participants
product interactions in a combinatory recommendation will
were randomly assigned to one of two recommendation condi-
make the recommendation a more diagnostic signal. In other
tions (combinatory vs. substitute) using a between-subjects
words, explicitly demonstrating an understanding of relation-
design.
ships among multiple products will signal greater expertise.
All participants were asked to imagine having some empty
wall space in their bedroom that they wanted to fill with a
piece of art. This empty wall space could accommodate, at
most, one 18′′ × 24′′ piece of art. Participants then imagined Method
shopping at a store to find a piece of 18′′ × 24′′ art. While Participants from MTurk (N = 301) completed the study for a
they were shopping, the store’s interior design adviser provided nominal payment. Nine responses from duplicate IDs were
them with a recommendation. Similar to Study 1b, in both con- excluded, leaving a sample of 292 responses (Mage = 41
ditions, the adviser discussed two pieces of wall art, one depict- years, 52% female). Participants were randomly assigned to
ing the Isle of Skye and one depicting the Botallack Tin Mine. one of two combinatory recommendation conditions (explicit
Both pieces of art measured 18′′ × 24′′ . Thus, the participant vs. implicit) using a between-subjects design.
would only be able to fit one of these pieces in their Participants imagined dining at a restaurant, where their
bedroom. In the combinatory recommendation condition, the waiter offered a three-course meal recommendation. In the
adviser noted how the two pieces should be hung together. In implicit condition, the waiter emphasized each dish’s individual
contrast, in the substitute condition, the adviser noted that the features, recommending the spring salad, followed by the king
participant could hang either of these pieces on their empty salmon, and finally a chocolate and strawberry crepe dessert. In
wall. the explicit condition, the waiter made the same recommenda-
Participants then rated their perceived expertise of the inte- tion but also explicitly mentioned that the three dishes go
rior designer using the three-item scale used in our previous well together, putting emphasis on the interactions across the
studies (α = .94). In addition, they rated how helpful the interior dishes. Thus, although both conditions offered the same combi-
designer was in finding art to fill the participant’s empty wall natory recommendation, resulting in the same meal experience,
space (1 = “Not at all,” and 9 = “Very”). the waiter in the explicit condition exhibited more knowledge
162 Journal of Marketing Research 60(1)

about the relationships among the dishes. Participants then rated fill downtime at an event. In both joint usage conditions, the
the waiter’s expertise using the three-item scale used in our pre- DJ recommended three contemporary songs to play at an
vious studies (α = .94). event because they have rhythms that complement one
another. However, in the close joint consumption condition,
the DJ stated that the transitions between the songs (i.e., the
Results end of one song and the beginning of the next) could be
A one-way ANOVA revealed a significant effect of the com- blended. Thus, the three songs would be consumed immedi-
binatory recommendation on perceived expertise, where per- ately after one another as a combined entity. In contrast, in
ceptions of expertise were greater in the explicit condition the distant joint usage condition, the DJ stated that each song
(M = 6.90) than in the implicit condition (M = 5.90; F(1, could be played at different points of downtime at the event.
290) = 34.56, p < .001, ω2p = .103). Therefore, each song would be consumed separately, but
within the same overall consumption occasion (i.e., the
event). In the substitute condition, the DJ recommended
Discussion playing any one of the three contemporary songs during down-
In support of H2, we find that even in situations where multiple time because the rhythms of the songs were good alternatives.
products are expected to be consumed together, explicitly Participants then rated their perceived expertise of the DJ using
acknowledging the compatibility of these products can be the three-item scale used in our previous studies (α = .96). They
more effective in signaling expertise. In the next study, we also rated the extent to which they expected that the songs in
test H3 and establish a boundary condition for when a combina- the recommendation would interact with one another (1 = “Not
tory recommendation signals expertise. at all,” and 9 = “Very much”).

Results
Study 3b: Greater Temporal Distance
Between Consumption Events Reduces The songs in the close joint usage condition (M = 7.81) were
perceived as interacting significantly more than those in the
Perceptions of Expertise
distant joint usage condition (M = 7.01; F(1, 294) = 10.05,
Although products in a combinatory recommendation are p = .002, ω2p = .030) and substitute condition (M = 6.29;
expected to be jointly used during a single consumption occa- F(1, 294) = 36.88, p < .001, ω2p = .108), confirming the effec-
sion, what qualifies as a consumption occasion might vary tiveness of our manipulation.
across different situations. We expect this to affect the nature A one-way ANOVA revealed a significant effect of recom-
of the interaction between product attributes and, ultimately, mendation on perceived expertise (F(2, 294) = 5.40, p = .005,
the effectiveness of a combinatory recommendation. For ω2p = .029). The close joint usage recommendation (M = 7.45)
instance, the time between when each individual product is con- signaled significantly more expertise than a distant joint usage
sumed (i.e., the individual consumption act) within the con- recommendation (M = 6.76; F(1, 294) = 10.70, p = .001, ω2p =
sumption occasion can vary. On one extreme, the products .032). Thus, recommending shorter temporal intervals
recommended could be consumed with the individual acts between the songs signaled greater expertise.
spread out in time. On the other, the products could be con- Consistent with our previous findings, the close joint usage
sumed jointly or in close succession. Study 3b tests whether recommendation also signaled marginally more expertise than
the temporal distance between the individual consumption the substitute recommendation (M = 7.06; F(1, 294) = 3.58,
acts of the products included in a combinatory recommendation p = .059, ω2p = .009). Interestingly, there was no significant dif-
affects perceptions of expertise. We reason that the shorter the ference between the distant joint usage and substitute recom-
distance between consumption acts, the more consumers expect mendation conditions (F = 1.94, p = .165). It thus appears that
interactions between product attributes. Thus, we predict that recommending the joint consumption of products where the
the shorter the temporal intervals suggested in a combinatory consumption acts are significantly spaced apart might function
recommendation, the more an adviser signals expertise. as a boundary of combinatory recommendations’ advantage in
signaling expertise. This may be because there is more ambigu-
ity about the advisers’ understanding of interactions or because
Method the depth of knowledge required to understand product interac-
Participants from MTurk (N = 301) completed the study for a tions in distant joint usage occasions is modest.
nominal payment. Four responses from duplicate IDs were
excluded, leaving a sample of 297 responses (Mage = 41
years, 52% female). Participants were randomly assigned to Discussion
one of three recommendation conditions (close joint usage, In support of H3, we find that temporal distance between con-
distant joint usage, substitute) using a between-subjects design. sumption acts in a combinatory recommendation affects con-
Participants imagined striking up a conversation with a DJ sumers’ perceived expertise of the adviser. We conceptually
who offered a recommendation about what music to play to replicate our findings by manipulating spatial distance (see
D’Angelo and Valsesia 163

Appendix for full study write-up). An adviser’s combinatory is .86, the product attitude AVE is .79, and the squared correla-
recommendation that pairs products consumed close in spatial tion between the two variables is .33, meeting the Fornell and
distance (i.e., pairing a pindot tie with an oxford shirt) signals Larcker (1981) criterion for discriminant validity.
greater expertise than one that pairs products farther in spatial Participants then responded to measures of potential alterna-
distance (i.e., pairing a pindot tie with a pair of oxford socks; tive drivers of product attitudes. They rated to what extent the
F(1, 195) = 6.97, p = .009, ω2p = .029). interior design adviser has good taste when it comes to art on
Notably, in this study we also introduce a boundary condi- a seven-point scale (1 = “Not at all,” and 7 = “Very much”)
tion for combinatory recommendations, where larger temporal and to what extent they perceived the art to be versatile on a
intervals between consumption acts may attenuate a combina- seven-point scale (1 = “Not at all,” and 7 = “Very much”).
tory recommendation’s ability to signal expertise above and Finally, participants rated to what extent they would describe
beyond a substitute recommendation. themselves as an interior design expert (1 = “Not at all,” and
The next studies investigate the practical importance of a 7 = “Very much”).
combinatory recommendation by examining its impact on con-
sumers’ product attitudes and behavior.
Results
The Isle of Skye wall art was rated more positively in the com-
Study 4: Combinatory Recommendations
binatory recommendation condition (M = 7.35) than in the sub-
Improve Product Attitudes via Expertise stitute recommendation condition (M = 6.80; F(1, 190) = 7.86,
Signals p = .006, ω2p = .034). The interior designer in the combinatory
Study 4 explores several important questions. First, we test recommendation condition was perceived as having greater
whether a combinatory recommendation affects consumers’ expertise (M = 7.75) than in the substitute condition (M =
attitudes toward products included in the recommendation. 6.55; F(1, 190) = 30.69, p < .001, ω2p = .134). There was no dif-
Second, we examine whether the effect of combinatory recom- ference between conditions for taste (F(1, 190) = 1.09, p = .297)
mendations on product attitudes is driven by perceptions of and versatility (F < 1).
expertise or other alternative explanations, namely perceptions Importantly, a mediation analysis (10,000 bootstraps;
that the recommender innately has good taste (rather than PROCESS Model 4; Hayes 2017) was conducted, testing
knowledge-based expertise) or that making a combinatory rec- expertise, taste, and versatility as potential drivers of product
ommendation highlights the versatility of the products included attitudes. The analysis revealed that the effect of the combi-
in the recommendation. Finally, this study examines whether natory recommendation on wall art attitude was driven by
one’s own expertise plays a role in the effect of combinatory perceived expertise of the adviser (indirect effect: b = .23,
recommendations on product attitudes. Although consumers SE = .09, 95% confidence interval [CI]: [.07, .41])2 but not
may often perceive expertise from an individual’s combinatory taste (indirect effect: b = .11, SE = .11, 95% CI: [−.09, .34])
recommendation, this expertise signal may be weakened if con- or versatility (indirect effect: b = .00, SE = .02, 95% CI:
sumers themselves have knowledge of product interactions. For [−.04, .04]). Thus, using a combinatory recommendation
these consumers, an individual’s combinatory recommendation increased perceptions of expertise of the adviser, which in
could be considered less diagnostic of their unique expertise. turn, increased attitudes toward the wall art included in the
combinatory recommendation.
We also analyzed whether the effect of combinatory recom-
Method mendations on perceived expertise of the adviser depended on
one’s own self-perceived expertise in interior design (10,000
Participants from MTurk (N = 200) completed the study for a
bootstraps; PROCESS Model 7; Hayes 2017). The interaction
nominal payment. Eight responses from duplicate IDs were
of recommendation and self-perceived expertise on perceived
excluded, leaving a sample of 192 responses (Mage = 41
expertise of the adviser was marginally significant (F(1, 188)
years, 60% female).
= 3.24, p = .073). Furthermore, the Johnson–Neyman technique
Participants were randomly assigned to one of two recom-
(Spiller et al. 2013) revealed that a combinatory recommen-
mendation conditions (combinatory vs. substitute) using a
dation significantly increased perceptions of the adviser’s
between-subjects design. We used the same stimuli from
expertise for values of self-perceived expertise less than
Study 1b but collected different measures. Participants rated
4.32 (bJN = .70, SE = .35, t = 1.97, p = .05), which represents
their attitudes toward the Isle of Skye art using three items
90% of our sample. That is, the combinatory recommendation
(how positive are your impressions of, how much do you
had a stronger effect on perceptions of the adviser’s expertise
like, and how likely are you to try out the Isle of Skye art) on
among those who had less expertise in interior design. The
a nine-point scale (1 = “Not at all,” and 9 = “Very much”).
index of moderated mediation was not significant at the 95%
The three items were averaged into one composite measure of
product attitude (α = .90). Participants also rated their perceived
expertise of the interior designer using the three-item scale used 2
The direct effect was not significant (t = −.27, p = .786), indicating full medi-
in our previous studies (α = .97). The perceived expertise AVE ation (Zhao, Lynch, and Chen 2010).
164 Journal of Marketing Research 60(1)

confidence level (index = −.143, 95% CI: [−.32, .02]) but was procedure (https://aspredicted.org/7XX_NGZ), we collected
significant at the 90% confidence level (90% CI: [−.29, −.01]). responses from unique IDs until we reached our desired
sample of 500 complete responses (Mage = 36 years, 56%
female). Participants were randomly assigned to one of two rec-
Discussion ommendation conditions (combinatory vs. substitute).
In Study 4, we conceptually replicate our finding that a combi- In both conditions, participants read three Instagram posts
natory recommendation signals expertise. Furthermore, we find from a food influencer. Each post recommended two
that this signal of expertise, in turn, can increase consumers’ food-related items. In the combinatory condition, the influencer
attitudes toward the products included in the combinatory rec- recommended that these two items be paired together (e.g., one
ommendation, supporting H4. At the same time, we address post recommended pairing apples and kiwifruit for a healthy
alternative explanations of taste and product versatility. It snack; see the Web Appendix for all posts). In the substitute
appears that combinatory recommendations do not affect per- condition, the influencer recommended that these items could
ceptions that the recommender innately has good taste nor be good alternatives (e.g., one post recommended eating
change how versatile the products included in the recommenda- apples or kiwifruit for a healthy snack).
tion are expected to be. Participants rated the influencer’s food expertise using the
Although these results do not reach full statistical signifi- same three item-measure from Study 1a (α = .96). Next, partic-
cance, we also find that the effect of combinatory recommenda- ipants across conditions viewed the same fourth post from the
tions on expertise is somewhat weaker among those who influencer, where the influencer recommended a type of
perceive themselves to have expertise in that category. It may cookie with no recommendation of a compatible or substitute
be that those who have prior knowledge in a category are less item. Thus, there was no cue of additional qualities of this
likely to consider external cues when making judgments cookie in either condition. Participants then rated their attitude
(Miller and Curry 2013). toward this brand of cookie using the same product attitude
In the next study, we extend the findings of Study 4 to measure used in Study 4 (α = .90). The perceived expertise
another important type of adviser, social media influencers, AVE is .83, the product attitude AVE is .75, and the squared
and show that the persuasive effects of combinatory recommen- correlation between the two variables is .38, suggesting dis-
dations extend beyond consumers’ attitudes toward the prod- criminant validity.
ucts included in the recommendation.

Results
Study 5: Expertise Signals from Combinatory
The influencer signaled greater expertise when using a combi-
Recommendations Improve Subsequent natory recommendation (M = 6.93) than when using a substi-
Product Recommendations tute recommendation (M = 5.44; F(1, 498) = 100.15, p < .001,
Although we rule out the role of versatility in Study 4, in ω2p = .17).
Study 5 we tackle the broader question of whether the effect Product attitude toward the cookie was also rated more
of combinatory recommendations on product attitudes positively after the influencer used the combinatory recom-
depends on consumers making different inferences about the mendation (M = 6.57), compared with the substitute recom-
qualities of the items recommended. If consumers’ product atti- mendation (M = 5.95; F(1, 498) = 18.87, p < .001, ω2p =
tudes improve because combinatory recommendations affect .035). Importantly, a mediation analysis (10,000 bootstraps;
perceptions of a product’s qualities, a combinatory recommen- PROCESS Model 4) revealed that the effect of recommenda-
dation should not improve consumers’ attitudes toward prod- tion on product attitude was driven by perceived expertise of
ucts not included in the recommendation. However, if the influencer (indirect effect: b = .86, SE = .10, 95% CI:
expertise signals are indeed a driver of improved product atti- [.67, 1.06]).3
tudes, consumers’ attitudes toward subsequent recommenda-
tions not included in the original combinatory
recommendation should improve. That is, once a person has Discussion
established category expertise via a combinatory recommenda-
This study provides additional evidence that a combinatory rec-
tion, their subsequent product recommendations in that cate-
ommendation can signal expertise and that this expertise signal
gory, even those not included in a combinatory
can subsequently improve consumers’ attitudes, not just toward
recommendation, should be viewed more favorably. This
the products included in the combinatory recommendation, but
study tests this prediction.
also toward products included in a subsequent recommenda-
tion. Next, we use a field study measuring real behavior to dem-
Method onstrate how combinatory recommendations can benefit firms.

We recruited Instagram users from MTurk who completed the


study for a nominal payment. Following a preregistered 3
The direct effect was marginally significant (t = −1.92, p = .055).
D’Angelo and Valsesia 165

Study 6: Combinatory Recommendations product recommendations, but the quote did not mention
Influence Real Behavior whether the products go well together in a skin-care routine
(see the Web Appendix for ads used).
In Study 6, we partnered with a skin-care company that hosts an Furthermore, a caption appeared below the carousel card
online skin-care advice platform. The company specifically image. In the ad with the combinatory recommendation
asks skin-care advice givers to post recommendations of multi- present, the caption emphasized how the advice givers on the
ple skin-care products that work well together in a skin-care skin-care platform could help consumers find the best combina-
routine (i.e., provide combinatory recommendations). The tion of products for their skin. In the ad without the combinatory
company believes it is uniquely positioned in the marketplace recommendation, the caption emphasized how the advice givers
with its use of combinatory recommendations since most on the platform could help consumers find the best types of
other beauty-related review platforms only provide recommen- products for their skin.
dations for individual skin-care products. Using an Instagram In both conditions, when an Instagram user clicked on the
A/B test, Study 6 examines whether consumers will find skin- ad, they were directed to the same front page of the skin-care
care advertising featuring combinatory recommendations from platform website. Following Paharia (2020), for our statistical
advice givers more appealing than similar advertising where analyses, we used the number of impressions for each ad as
advice givers recommend the same products without combina- the N. We examined three distinct dependent variables: click-
tory recommendations. From the findings in our previous through rate (CTR), sign-up rate, and cost per click. These var-
studies, we reason that skin-care advice givers who use combi- iables are described subsequently.
natory recommendations signal greater expertise. Thus, we
predict that consumers will be more likely to visit the skin-care Click-through rate. We measured the number of clicks on each
advice platform knowing that the advice givers on that platform ad and used the CTR (number of ad clicks divided by number
have expertise. of impressions) as the key DV.

Sign-up rate. In each ad, URL parameters were added to allow


Method us to capture how many Instagram users, among those who
Facebook Ad Manager (which places advertisements on clicked the ad, ended up signing up for the skin-care advice
Instagram, Facebook, Facebook Messenger, and other plat- platform. We used sign-up rate (number of sign-ups divided
forms) has an A/B split test feature, which enables marketers by number of ad clicks) as another measure of analysis.
to compare the effectiveness of different advertisements. We
used Facebook Ad Manager to run ads on Instagram only. Cost per click. Finally, we examined the average cost of an ad
We created advertisements with the following settings: split click (number of ad clicks divided by total dollar amount
test on creative; four-day test;4 age 18 years and up; women; spent). A lower cost per click implies a more effective ad
location: United States; maximum daily budget: $66;5 interests: campaign.
cosmetics, beauty, skin care, self-care; optimization: link clicks;
bid strategy: lowest cost; placement: Instagram; charged: cost Results
per click. We chose the carousel format, in which the ad con-
sists of multiple images a consumer can swipe through, Click-through rate. A logistic regression predicting CTR
because Facebook recommended this format to increase click- revealed that the ad with the combinatory recommendation
through rates (Facebook for Business 2015). These settings present led to significantly more clicks (1.52%, n = 43,331)
hold all elements constant except for the messaging in the than the ad without the combinatory recommendation (1.26%,
advertisement. n = 36,966; b = −.19, SE = .06, Wald χ2(1) = 9.92, p = .002).
Each ad version (combinatory recommendation: present vs. The same statistical conclusions apply for CTR when we
absent) consisted of three carousel cards (i.e., three images). In used reach, instead of impressions, as our N (see the Web
the ad with the combinatory recommendation present, each card Appendix for supplementary analysis).
quoted and depicted one skin-care advice giver’s recommenda-
tion for which products go well together in a skin-care routine. Sign-up rate. Among those who clicked on the ad to visit the
In the ad without the combinatory recommendation, each card skin-care website, the sign-up rate was greater among those
quoted and depicted the same skin-care advice giver’s who came from the ad with the combinatory recommendation
(7.12%; 47 sign-ups, n = 660) than among those who came
from the ad without the combinatory recommendation
4
Facebook recommends a four-day test for the most reliable results when con- (4.51%; 21 sign-ups, n = 466; b = −.49, SE = .27, Wald χ2(1)
ducting a split test (Facebook 2020).
5
= 3.24, p = .072). Although this difference is marginally signif-
The $66/day budget ($264 total) is split evenly across two versions of the
icant, the effect of the combinatory recommendation on the
advertisement. When we entered our desired settings for the split test, $66
was computed as Facebook’s daily budget recommendation for conducting a sign-up rate is still notable, as this effect may have been weak-
split test with at least 80% power. Our actual expense came in slightly under ened from users in both conditions landing on the same home
budget at $257.17. page featuring combinatory recommendations.
166 Journal of Marketing Research 60(1)

Cost per click. Finally, in terms of cost per click, the ad with the nature of the combination recommended (the extent to which
combinatory recommendation (CPC = $.19) outperformed the products are expected to interact) can affect the effectiveness
ad without the combinatory recommendation (CPC = $.28). of combinatory recommendations as signals. We also find
some evidence that one’s own knowledge in a product category
may weaken the adviser’s expertise signal. To further explore
Discussion this possibility, we ran an additional study testing whether rec-
ommendation conventionality affects perceived expertise. We
This Instagram field experiment demonstrates the external
find that at high and low levels of conventionality, consumers
validity of our findings, measuring real consumer behavior.
feel more knowledgeable about how products will interact
We note that given Facebook’s proprietary algorithms, the ad
with one another. Thus, consumers perceive less expertise for
optimization strategy remains a black box. Therefore, if a par-
these kinds of recommendations. For example, recommending
ticular advertisement performs better on an A/B test, it is
a combination of lemonade with iced tea (also known as an
unclear whether the advertisement performed better overall or
Arnold Palmer), which consumers already know interact well
whether it optimized better (see Hardisty and Weber [2020]
with each other, will not be as strong of a signal of expertise.
for an in-depth discussion). However, our experimental
At the same time, recommending a mixture of lemonade with
methods from Studies 4 and 5, coupled with this field study,
milk, which consumers anticipate will not interact well with
suggest that combinatory recommendations can be effective
each other, will also not signal expertise. However, at moderate
signals of expertise, which subsequently lead to important
levels of conventionality, consumers feel less certain about the
downstream consequences, including improved product atti-
interaction outcome and thus are more likely to see combinatory
tudes and website traffic.
recommendations as signals of expertise (see the Web
Appendix for full study write-up).
Our work makes important contributions to several streams
General Discussion of literature. For one, we add to the literature on signaling by
In recent years, retailers have made significant investments in identifying a cue that signals expertise and renders more persua-
hiring individuals (e.g., stylists, designers, chefs, sponsored sive recommendations. Importantly, the cue we identify may be
social media influencers) to advise consumers on what to pur- more readily actionable than some of the cues identified in prior
chase across a number of product domains, from food to cloth- work. For instance, projecting a confident attitude might be
ing to interior design. These consumption advisers want to natural to some but might be very hard for others.
establish themselves as trusted experts in their field, particularly Our work also extends the literature on persuasion in two
as the number of advisers available to consumers proliferates. important ways. First, we identify a novel persuasive tactic
This research identifies combinatory recommendations as a and explore when and why this tactic is effective. Second,
novel cue advisers can use to signal category expertise to con- Hardesty, Bearden, and Carlson (2007) find that consumers’
sumers. Across nine studies, we find that using combinatory persuasion knowledge may be activated when marketers
recommendations has a positive effect on expertise perceptions. bundle multiple items together to increase revenue over what
When advisers make a combinatory recommendation, they would have been obtained had the products been priced sepa-
display their understanding of how product attributes interact rately. Thus, one could reason that consumers perceive an
and the outcome of such interactions. Thus, keeping the adviser’s combinatory recommendation as an upselling tactic,
number of items recommended constant, combinatory recom- ultimately undermining the persuasiveness of the adviser.
mendations can increase perceived depth (but not breadth) of Interestingly, we find that this is not the case, as a combinatory
knowledge in a product category. recommendation generally improves, rather than undermines,
Signals of advisers’ expertise matter, as these signals an adviser’s persuasiveness.
improve consumers’ attitudes toward the products recom- Importantly, we also contribute to the bundling literature
mended by these advisers. This effect includes products in the and the nascent literature on curation. Prior work predomi-
combinatory recommendation and extends to other, subse- nantly focuses on evaluating the products included in the
quently recommended, products that were not part of a combi- bundle, but not the perceptions of the curator of bundle. We
natory recommendation. offer a new contributory explanation for the well-established
Although we find that the effects identified generally hold finding that bundling complementary products improves con-
across different types of combinations and interactions (e.g., sumers’ product attitudes. That is, we show that these
blended combinations, such as drinks; temporal combinations, improved attitudes may be driven by the bundle curator
such as skin-care routines), the extent to which a combinatory being perceived as an expert. Further, whereas most work
recommendation can be an effective signal of expertise can on product complementarity examines the pairing of true com-
depend on a variety of factors. Our theorizing suggests that plements, like chips and salsa (Huh, Vosgerau, and
combinatory recommendations work as expertise signals Morewedge 2016; Karataş and Gürhan-Canli 2020; Rahinel
because they highlight one’s ability to understand attribute and Redden 2013), we examine products that could be per-
interactions. Thus, both the understanding of the interaction dis- ceived as either complements or substitutes on the basis of
played in the recommendation (implicit versus explicit) and the recommendation type. We show that simply framing
D’Angelo and Valsesia 167

products as complements can also yield improved product process interactions among products. Would this signal
attitudes. greater expertise and lead to other positive outcomes, including
greater engagement with the adviser? We find initial evidence
to support this prediction from Smart Closet, an online platform
Managerial Implications where users post clothing pieces that one could wear together.
Our work also informs both retailers and individual advisers of An analysis of the nonduplicate posts displayed on the first
the effectiveness of using combinatory recommendations to page (N = 399) reveals a significant correlation between the
establish one’s expertise. Retailers should consider training number of clothing pieces in the post and the number of likes
advisers, be they sponsored influencers or salespeople, to use the post received (r = .14, p = .004). Future research could
combinatory recommendations to improve perceptions of exper- also explore whether there are limits to the number of compat-
tise and, ultimately, attitudes toward recommended products. At ible items included in a recommendation a consumer finds
the same time, retailers should emphasize these advisers’ ability believable.
to create combinations through their marketing communications. Although we examined compatibility among multiple prod-
For instance, Nordstrom’s personal styling service, Trunk Club, ucts, another direction for future research might explore combi-
simply lists “Advice putting outfits together” as one of six ser- natory recommendations that offer opinions about compatibility
vices offered by its stylists. Nordstrom could consider prioritiz- among a product and the traits of the consumer. For example,
ing this aspect of the service in its marketing communications does recommending whether a skin-care product goes well
as a way of signaling the expertise of its stylists. Similarly, indi- with oily skin or whether a dress complements a pear-shaped
viduals, such as social media influencers trying to break through body signal expertise? These types of combinatory recommen-
the clutter, may consider using combinatory recommendations to dations may require a more complex processing of relationships
position themselves as category experts. between the nonalignable attributes of the product and the con-
These managerial recommendations come with an important sumer. Thus, one could reason that these types of combinatory
caveat. We have learned that not all combinatory recommenda- recommendations might signal even greater expertise than the
tions function in the same way, and that both the language used combinatory recommendations explored in this research.
in making the recommendation and the expected interactions One might be concerned that recommending the joint usage
between products recommended matter, highlighting that com- of two or more products might be perceived as an upselling
binatory recommendations actually need to display depth of technique, triggering consumers’ persuasion knowledge and
knowledge in order to be effective signals of expertise. skepticism (Friestad and Wright 1994) and leading to lower
Our research also bears potentially important managerial purchase intentions. Interestingly, we find no evidence of
insights beyond the type of consumption advice studied in these potential negative consequences in our work. It is none-
this work. For instance, retailers could consider using direct theless possible that, under certain situations, a combinatory
mail and personalized communication to provide combinatory recommendation might indeed trigger a negative reaction in
recommendations. In this vein, fashion retailer Mango sends consumers. If and when this could happen may be another
follow-up emails that contain suggestions of clothing pieces potentially interesting question for future research.
consumers could wear their recent purchase with. Similarly, Finally, there is growing interest in how consumers interact
review platforms could prompt combinatory recommendations with artificial intelligence (AI) and the extent to which percep-
from consumers. A clothing retailer could ask consumers to fill tions of people extend to perceptions of AI agents. Future
out the question “What did you wear this with?” when submit- research might thus explore whether AI recommendation
ting a review. Online clothing retailers could also encourage agents can signal expertise using the same tactics as human
consumers to post outfit photos, a practice already implemented advisers. These future directions for research will help
by retailers like Rent the Runway and Old Navy. That way, uncover additional conditions in which combinatory recom-
other consumers can get a better idea of which clothing items mendations are most likely to signal expertise.
are compatible with each product. Our findings from Study 6
suggest that these types of recommendations should generate
consumer interest and differentiate the online platform. Appendix: Greater Spatial Distance Between
Products in a Combinatory Recommendation
Future Research Reduces Perceptions of Expertise
Future research could examine different compositions of com-
binatory recommendations. In our studies, we intentionally This study tests whether spatial distance between the products in
held the number of products in our combinatory and substitute a combinatory recommendation affects perceptions of expertise.
recommendations constant to control for category breadth.
However, future research could compare combinatory recom-
mendations involving varying numbers of products.
Method
According to our theorizing, a person who can recommend Participants from MTurk (N = 201) completed the study for a
many compatible products may exhibit a deeper ability to nominal payment. Four responses from duplicate IDs were
168 Journal of Marketing Research 60(1)

excluded, leaving a sample of 197 responses (Mage = 37 years, Babin, Jessica and John Hulland (2019), “Exploring Online Consumer
49% female). Participants were randomly assigned to one of Curation as User-Generated Content,” Spanish Journal of
two combinatory recommendation conditions (close vs. Marketing–ESIC, 23 (3), 325–37.
distant) using a between-subjects design. Bellezza, Silvia, Francesca Gino, and Anat Keinan (2014), “The Red
Participants imagined that a sales associate provided a com- Sneakers Effect: Inferring Status and Competence from Signals
binatory recommendation. In the close condition, the sales asso- of Nonconformity,” Journal of Consumer Research, 41 (1), 35–54.
ciate recommended pairing a pindot tie with an oxford shirt Bobb, Brooke (2019), “Karla Welch, Stylist to Megan Rapinoe,
because they go well together. In the distant condition, the Tracee Ellis Ross, and More, Launched an Online Styling
sales associate recommended pairing a pindot tie with a pair App,” Vogue (September 25), https://www.vogue.com/article/
of oxford socks. karla-welch-wishi-app-personal-stylist.
Participants then completed our three-item measure of per- Chandler, Jesse J. and Gabriele Paolacci (2017), “Lie for a Dime:
ceived expertise by rating the expertise of the sales associate When Most Prescreening Responses Are Honest but Most Study
(1 = “Not at all,” and 9 = “Very much”; α = .94). Participants Participants Are Impostors,” Social Psychological and
also rated the spatial distance between the products recom- Personality Science, 8 (5), 500–508.
mended (1 = “The pieces are worn very far apart,” and 9 = Clarkson, Joshua J., Chris Janiszewski, and Melissa D. Cinelli (2013),
“The pieces are worn very close together”). “The Desire for Consumption Knowledge,” Journal of Consumer
Research, 39 (6), 1313–29.
Coker, James (2020), “Covid-19: John Lewis Launches Virtual
Results and Discussion Consultations for Customers,” Essential Retail (April 14), https://
www.essentialretail.com/news/covid-19-john-lewis-virtual/.
Confirming the effectiveness of our manipulation, the clothing
D’Angelo, Jennifer K., Kristin Diehl, and Lisa A. Cavanaugh (2019),
pieces in the close condition were perceived as closer (M =
“Lead by Example? Custom-Made Examples Created by Close
7.95) than those in the distant condition (M = 3.84; F(1, 195)
Others Lead Consumers to Make Dissimilar Choices,” Journal of
= 225.95, p < .001, ω2p = .533).
Consumer Research, 46 (4), 750–73.
A one-way ANOVA revealed a significant effect of combi-
Davenport, Thomas, Abhijit Guha, Dhruv Grewal, and Timna Bressgott
natory recommendation on perceived expertise, where percep-
(2020), “How Artificial Intelligence Will Change the Future of
tions of expertise were greater in the close condition (M =
Marketing,” Journal of the Academy of Marketing Science, 48 (1), 24–42.
7.23) than in the distant condition (M = 6.72; F(1, 195) =
Deng, Xiaoyan, Sam K. Hui, and J. Wesley Hutchinson (2010),
6.97, p = .009, ω2p = .029).
“Consumer Preferences for Color Combinations: An Empirical
These results suggest that products consumed more closely
Analysis of Similarity-Based Color Relationships,” Journal of
in terms of spatial distance, such that the interaction of the prod-
Consumer Psychology, 20 (4), 476–84.
ucts is likely more intertwined, can amplify the effectiveness of
Diehl, Kristin, Erica Van Herpen, and Cait Lamberton (2015),
a combinatory recommendation in signaling expertise.
“Organizing Products with Complements Versus Substitutes:
Effects on Store Preferences as a Function of Effort and
Associate Editor Assortment Perceptions,” Journal of Retailing, 91 (1), 1–18.
James Bettman Facebook (2020), “Best Practices for A/B Testing,” (accessed March 22,
2020), https://www.facebook.com/business/help/290009911394576.
Author Note Facebook for Business (2015), “Improving Ad Performance with the
Carousel Format,” (accessed March 22, 2020), https://www.
The authors contributed equally and are listed in alphabetical order.
facebook.com/business/news/carousel-ads.
Fornell, Claes and David F. Larcker (1981), “Structural Equation Models
Declaration of Conflicting Interests with Unobservable Variables and Measurement Error: Algebra and
The author(s) declared no potential conflicts of interest with respect to Statistics,” Journal of Marketing Research, 18 (3), 382–88.
the research, authorship, and/or publication of this article. Friestad, Marian and Peter Wright (1994), “The Persuasion Knowledge
Model: How People Cope with Persuasion Attempts,” Journal of
Funding Consumer Research, 21 (1), 1–31.
Gershoff, Andrew D., Susan M. Broniarczyk, and Patricia M. West (2001),
The author(s) received no financial support for the research, author-
“Recommendation or Evaluation? Task Sensitivity in Information
ship, and/or publication of this article.
Source Selection,” Journal of Consumer Research, 28 (3), 418–38.
Glenday, John (2020), “Three-Quarters of Brands Have Upped
References Influencer Spend Despite Pandemic,” The Drum (October 1),
Alba, Joseph W. and J. Wesley Hutchinson (1987), “Dimensions of https://www.thedrum.com/news/2020/10/01/three-quarters-brands-
Consumer Expertise,” Journal of Consumer Research, 13 (4), 411–54. have-upped-influencer-spend-despite-pandemic.
Aribarg, Anocha and Natasha Zhang Foutz (2009), “Category-Based Gorin, Amy (2018), “The Rise of the Supermarket Nutritionist,” Food
Screening in Choice of Complementary Products,” Journal of Network, https://www.foodnetwork.com/healthyeats/news/2018/
Marketing Research, 46 (4), 518–30. supermarket-rds.
D’Angelo and Valsesia 169

Guiltinan, Joseph P. (1987), “The Price Bundling of Services: A Petty, Richard E. and John T. Cacioppo (1986), “The Elaboration
Normative Framework,” Journal of Marketing, 51 (2), 74–85. Likelihood Model of Persuasion,” in Communication and
Hardesty, David M., William O. Bearden, and Jay P. Carlson (2007), Persuasion. New York: Springer, 1–24.
“Persuasion Knowledge and Consumer Reactions to Pricing Popkowski Leszczyc, Peter T.L. and Gerald Häubl (2010), “To Bundle
Tactics,” Journal of Retailing, 83 (2), 199–210. or Not to Bundle: Determinants of the Profitability of Multi-Item
Hardisty, David J. and Elke U. Weber (2020), “Impatience and Auctions,” Journal of Marketing, 74 (4), 110–24.
Savoring vs. Dread: Asymmetries in Anticipation Explain Price, Paul C. and Eric R. Stone (2004), “Intuitive Evaluation of
Consumer Time Preferences for Positive vs. Negative Events,” Likelihood Judgment Producers: Evidence for a Confidence
Journal of Consumer Psychology, 30 (4), 598–613. Heuristic,” Journal of Behavioral Decision Making, 17 (1), 39–57.
Hayes, Andrew F. (2017), Introduction to Mediation, Moderation, and Rahinel, Ryan and Joseph P. Redden (2013), “Brands as Product
Conditional Process Analysis: A Regression-Based Approach. Coordinators: Matching Brands Make Joint Consumption Experiences
New York: Guilford Press. More Enjoyable,” Journal of Consumer Research, 39 (6), 1290–99.
Hildebrand, Christian, Gerald Häubl, and Andreas Herrmann (2014), Sela, Aner, Liat Hadar, Sian Morgan, and Michal Maimaran (2019),
“Product Customization via Starting Solutions,” Journal of “Variety-Seeking and Perceived Expertise,” Journal of Consumer
Marketing Research, 51 (6), 707–25. Psychology, 29 (4), 671–79.
Howland, Daphne (2018), “Wayfair Draws up Interior Design Shocker, Allan D., Barry Bayus, and Namwoon Kim (2004), “Product
Services,” Retail Dive (August 16), https://www.retaildive.com/ Complements and Substitutes in the Real World: The Relevance of
news/wayfair-draws-up-interior-design-services/530237/. Other Products,” Journal of Marketing, 68 (1), 28–40.
Huh, Young Eun, Joachim Vosgerau, and Carey K. Morewedge Slovic, Paul and Sarah Lichtenstein (1971), “Comparison of Bayesian
(2016), “Selective Sensitization: Consuming a Food Activates a and Regression Approaches to the Study of Information Processing
Goal to Consume Its Complements,” Journal of Marketing in Judgment,” Organizational Behavior and Human Performance,
Research, 53 (6), 1034–49. 6 (6), 649–744.
Kallio, Eeva (2011), “Integrative Thinking Is the Key: An Evaluation Smith, Robert W. and Joseph P. Redden (2020), “The Role of Holistic
of Current Research into the Development of Adult Thinking,” Processing in Simultaneous Consumption,” Journal of Experimental
Theory & Psychology, 21 (6), 785–801. Social Psychology, 91, 104023.
Karataş , Mustafa and Zeynep Gürhan-Canli (2020), “When Sniezek, Janet A. and Timothy Buckley (1995), “Cueing and Cognitive
Consumers Prefer Bundles with Noncomplementary Items to Conflict in Judge-Advisor Decision Making,” Organizational
Bundles with Complementary Items: The Role of Mindset Behavior and Human Decision Processes, 62 (2), 159–74.
Abstraction,” Journal of Consumer Psychology, 30 (1), 24–39. Spence, Charles (2020a), “Food and Beverage Flavour Pairing: A
Libai, Barak, Eitan Muller, and Renana Peres (2013), “Decomposing the Critical Review of the Literature,” Food Research International,
Value of Word-of-Mouth Seeding Programs: Acceleration Versus 133 (July), 109124.
Expansion,” Journal of Marketing Research, 50 (2), 161–76. Spence, Charles (2020b), “Multisensory Flavour Perceptions:
Mannucci, Pier Vittorio and Kevyn Yong (2018), “The Differential Blending, Mixing, Fusion, and Pairing Within and Between the
Impact of Knowledge Depth and Knowledge Breadth on Senses,” Foods, 9 (4), 447–29.
Creativity over Individual Careers,” Academy of Management Spiller, Stephen A., Gavan J. Fitzsimons, John G. Lynch Jr., and Gary
Journal, 61 (5), 1741–63. H. McClelland (2013), “Spotlights, Floodlights, and the Magic
Miller, Banks and Brett Curry (2013), “Experts Judging Experts: The Number Zero: Simple Effects Tests in Moderated Regression,”
Role of Expertise in Reviewing Agency Decision Making,” Law & Journal of Marketing Research, 50 (2), 277–88.
Social Inquiry, 38 (1), 55–71. Tellis, Gerard J. (1986), “Beyond the Many Faces of Price: An Integration
Mitchell, Andrew A. and Peter A. Dacin (1996), “The Assessment of of Pricing Strategies,” Journal of Marketing, 50 (4), 146–60.
Alternative Measures of Consumer Expertise,” Journal of Valsesia, Francesca, Davide Proserpio, and Joseph C. Nunes (2020),
Consumer Research, 23 (3), 219–239. “The Positive Effect of Not Following Others on Social Media,”
Packard, Grant and David B. Wooten (2013), “Compensatory Journal of Marketing Research, 57 (6), 1152–68.
Knowledge Signaling in Consumer Word-of-Mouth,” Journal of Wilson, Marianne (2020), “Neiman Marcus Launches New Online Services,
Consumer Psychology, 23 (4), 434–50. Experiences,” Chain Store Age (July 23), https://chainstoreage.com/
Paharia, Neeru (2020), “Who Receives Credit or Blame? The Effects neiman-marcus-launches-new-online-services-experiences.
of Made-to-Order Production on Responses to Unethical and Yang, Defeng, Lu Jin, and Shibin Sheng (2017), “The Effect of
Ethical Company Production Practices,” Journal of Marketing, Knowledge Breadth and Depth on New Product Performance,”
84 (1), 88–104. International Journal of Market Research, 59 (4), 517–36.
Palmeira, Mauricio (2020), “Advice in the Presence of External Cues: Zhao, Min and Lan Xia (2020), “Joint or Separate? The Effect of
The Impact of Conflicting Judgments on Perceptions of Expertise,” Visual Presentation on Imagery and Product Evaluation,”
Organizational Behavior and Human Decision Processes, 156 International Journal of Research in Marketing, 38 (4), 935–52.
(January), 82–96. Zhao, Xinshu, John G. Lynch Jr., and Qimei Chen (2010),
Petre, Marian, Helen Sharp, and Jeffrey Johnson (2006), “Complexity “Reconsidering Baron and Kenny: Myths and Truths About
Through Combination: An Account of Knitwear Design,” Design Mediation Analysis,” Journal of Consumer Research, 37 (2),
Studies, 27 (2), 183–222. 197–206.

You might also like