You are on page 1of 9

Journal of Business Research 67 (2014) 2836–2844

Contents lists available at ScienceDirect

Journal of Business Research

New product adoption in social networks: Why direction matters


Oliver Hinz a,⁎, Christian Schulze b, Carsten Takac c
a
Technische Universität Darmstadt, Hochschulstr 1, 64289 Darmstadt, Germany
b
Frankfurt School of Finance & Management, Sonnemannstr 9-11, 60314 Frankfurt am Main, Germany
c
Goethe-University of Frankfurt, Grueneburgplatz 1, 60323 Frankfurt am Main, Germany

a r t i c l e i n f o a b s t r a c t

Available online 17 August 2012 Marketing managers and researchers generally agree that analyzing data from social networks and using
them to influence consumers' purchase decisions are useful strategies. However, not all social network
Keywords: data may identify the most influential customers. This empirical study of more than 300 students reveals
Social network analysis the low explanatory power of friendship networks (e.g., Facebook) and undirected-advice networks (e.g.,
Social contagion LinkedIn). Only directed-advice networks (e.g., Google+) clearly identify influential consumers. In addition,
Diffusion
the results challenge conventional wisdom that firms should target advisers assuming that they have the
Directed networks
Undirected networks
strongest influence on new product adoption. This study contradicts this common assumption and reveals
that structural equivalence drives product adoption more than cohesion because advisees' adoption pressures
advisers to purchase the product as well. Finally, the study shows the value of social network data beyond the
traditional ego-centric psychographic metrics, such as innovativeness or opinion leadership.
© 2012 Elsevier Inc. All rights reserved.

1. Introduction by other means (Schmitt, Skiera, & Van den Bulte, 2011; Trusov et
al., 2009; Villanueva, Yoo, & Hanssens, 2008). Furthermore, greater
Understanding new product adoption behavior is critical for both consumer activity on social network sites enables companies to gather
firms and researchers striving to explain and influence consumers' detailed information about users' social interactions and social relation-
decisions. Because influencing consumers through traditional adver- ships. Knowing who is connected to whom is instrumental to identify
tising seems to become less effective (Trusov, Bucklin, & Pauwels, and target influential consumers and to increase new product adoption
2009; Van den Bulte & Wuyts, 2007), firms continuously seek new (Bampo, Ewing, Mather, Stewart, & Wallace, 2008; Hinz et al., 2011).
ways to promote products and influence consumers' adoption deci- Marketing practitioners and researchers seem to agree that social
sions. According to sociological research, economic behavior embeds network data helps facilitate targeted marketing and influence con-
into the social environment (Granovetter, 1985). Social contagion af- sumers' new product adoption behavior. The presumed marketing
fects prospective customers, and some persons wield more influence potential of structural data from social network platforms such as
over purchase decisions than others (Godes & Mayzlin, 2009; Facebook and LinkedIn is visible in these firms' high market valua-
Goldenberg, Han, Lehmann, & Hong, 2009; Iyengar, Van den Bulte, & tions. Does information about the structure of consumers' social
Valente, 2011). Thus, more firms attempt to address individual con- networks from these platforms provide sufficient information to identify
sumers directly (Algesheimer, Borle, Dholakia, & Singh, 2010) and the customers who strongly influence other consumers' purchase deci-
base targeting strategies on consumers' social influence (Hinz, sions? This study addresses the fundamental research question: What
Skiera, Barrot, & Becker, 2011). For example, one tactic triggers and types of social networks enable firms to identify and target influential
influences consumer-to-consumer (C2C) communication (Hinz & consumers?
Spann, 2008; Libai et al., 2010). This approach seems promising due To determine social network data's suitability to identify influen-
to social network platforms' increasing popularity (e.g., Facebook, tial customers, the present research adopts two perspectives. First,
LinkedIn, or Google+). These platforms offer consumers ample oppor- this study examines the explanatory power of undirected networks
tunities for brand- and product-related discussions, increasing C2C for consumer product adoption and whether targeting individual
communication. consumers using information from undirected networks offers a
Recent research indicates that new customers influenced by C2C promising strategy for firms. In undirected networks, relationships
communication are more valuable to firms than customers acquired are mutual (e.g., Facebook, LinkedIn), without any observable rela-
tionship hierarchy (Scott, 2000). Conversely, directed networks
⁎ Corresponding author. Tel.: +49 6151 167 5220; fax: +49 6151 167 2220.
allow consumers to name another contact without requiring recipro-
E-mail addresses: hinz@wi.tu-darmstadt.de (O. Hinz), email@christian-schulze.de cation. Because directed networks (e.g., Google+) contain additional
(C. Schulze), carsten@takac.de (C. Takac). information about who talks and who listens, they might permit

0148-2963/$ – see front matter © 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jbusres.2012.07.005
O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844 2837

better identification of influential consumers and might help firms 2. Explaining new product adoption through social contagion and
target their marketing (Hinz et al., 2011). individual psychographic concepts
Second, this study investigates social contagion's effect and direc-
tion. Social networks' information seeking and learning models sug- 2.1. Social contagion
gest that not all social contacts are equal; consumers decide whom
to consult in specific situations (Borgatti & Cross, 2003). Consumer re- New product adoption is an imitation process (Mahajan & Muller,
lationships that revolve around a specific topic of common interest 1979; Rogers & Shoemaker, 1971). This theory applies to human be-
(e.g., cars, electronics, or fashion) may provide greater impact on pur- havior in general, which depends strongly on the social environment.
chase decisions in that domain than friendships do. This study com- For new product adoption, the social environment and social contacts
pares the usefulness of friendship networks (Facebook) and advice play important roles by providing and validating information about
networks, which either focus on a specific topic (e.g., professional re- the advantages and disadvantages of the product. Social contagion af-
lationships in LinkedIn) or permit various topical sub-networks (e.g., fects prospective adopters, and some people wield more influence
circles in Google +), to identify the most influential consumers. over adoption decisions than others (Godes & Mayzlin, 2009;
In addition to this comparison, this study investigates which con- Goldenberg et al., 2009; Iyengar et al., 2011). As Borgatti and Cross
sumers firms should target. Some research challenges long-accepted (2003) suggest, the ability to access and value another person's
truths about the importance of advisers or opinion leaders in product knowledge is a key factor in social information gathering and, ulti-
adoption processes (Van den Bulte & Stremersch, 2004; Watts & mately, social learning. Especially when mass media are omnipresent,
Dodds, 2007), suggesting that social contagion is better explained obtaining information is less problematic than finding trustworthy in-
by status considerations than by social learning under uncertainty. formation, and the challenge is to filter overwhelming data (Coleman,
Theory offers two competing explanations on how two consumers Katz, & Menzel, 1966).
(e.g., adviser vs. advisee) influence each other's product adoption de- For example, in a social network, friendship patterns, advice, com-
cisions. First, cohesion describes the popular view that an adviser munication, and support exist among social system members (Scott,
adopting a product likely exerts influence on an advisee who has 2000). Consumers base decisions on social cues (e.g., behavior of
not yet adopted the product. Second, structural equivalence suggests others) (Festinger, Schachter, & Back, 1950). Social networks gener-
that an advisee adopting a product exerts social pressure on an ad- ate trust, reduce uncertainty, and mitigate the information ambiguity
viser based on competitive concerns who has not yet adopted the prod- (Valente, 1996). Social contagion arises when proximate people use
uct, because the adviser fears losing his or her social status as adviser one another to manage the uncertainty of prospective adoption
(Burt, 1987). Research remains uncertain about which contagion effect (Granovetter, 1985). What causes people to behave similarly? In ad-
dominates (see Bowler, Dahlstrom, Seevers, & Skinner, 2011). If cohe- dition to homophily, sociology and network theory offer two plausi-
sion is dominant, firms should target well connected advisers and dis- ble explanations: cohesion and structural equivalence (Burt, 1987).
seminate information and influence many others to adopt as well. If “Cohesion” refers to socialization between adopter A and potential
social pressure arising from structural equivalence is dominant, firms adopter P. As social interaction increases between A and P, A's influ-
should target advisees instead, because these consumers' early adop- ence more likely triggers P's adoption. When A and P discuss an
tion behavior exerts pressure on well-connected advisers to adopt to open question, such as an adoption decision, both parties align their
preserve a certain social status (Bowler et al., 2011). This study investi- evaluations of the associated costs and benefits through strong com-
gates whether the effect of cohesion is stronger than the effect of struc- munication (Fischer, 1978). Structural equivalence reflects competi-
tural equivalence, or in other words, whether targeting advisers is tion mechanisms instead, so A and P need not engage in direct
indeed more effective than targeting advisees. social interaction. Their competing social positions in overlapping so-
Finally, this study analyzes the value of complete social network cial networks drive the adoption decision (Burt, 1987). For example,
information available from social network platforms compared to competition for a certain social position increases the social pressure
the value of ego-centric psychographic concepts (e.g., opinion leader- on P to adopt and avoid losing social position to A. Thus, structural
ship and innovativeness) (Katz & Lazarsfeld, 1955; Rogers, 1962, equivalence and cohesion provide two (potentially overlapping) con-
2003). From a theoretical perspective, the comparison sheds light cepts to explain social influence on new product adoption decisions
on the question whether social network information is a substitute through the bandwagon pressure of social contagion (Abrahamson
for or a complement to traditional ego-centric concepts. & Rosenkopf, 1997; Sailer, 1978).
This article offers both theoretical and practical perspectives. From a
theoretical perspective, systematically analyzing the explanatory power 2.2. Psychographic concepts
of two types of social networks (friendship vs. adviser–advisee) with
different degrees of information (undirected vs. directed) yields impor- Without information about a social networks' structure,
tant implications regarding the usefulness of different social networks researchers commonly use psychographic concepts, such as opinion
for social network analysis. As the empirical study shows, not all social leadership (e.g., Flynn, Goldsmith, & Eastman, 1994). Managers and
networks collect the same structural data and differ significantly in researchers consider opinion leadership important to new product
their explanatory power for targeted marketing. From a theory perspec- adoption decisions (Rogers, 1962, 2003). Opinion leadership plays a
tive, the article investigates the strength of competing explanations for central role in consumer behavior models and affects successful
social contagion: cohesion and structural equivalence. Again, the new product strategies (Flynn et al., 1994). The expansive notion
empirical study's results run counter to widely held beliefs about the describes the influence of some people on others' adoption decisions
processes underlying social influence among consumers. Managerially, (Katz & Lazarsfeld, 1955). Opinion leaders typically have greater
the article offers several important insights. Practitioners attempting to exposure to mass media, engage in more social experiences, and
influence consumers' new product adoption behavior based on social enjoy higher socio-economic status. As a reference point for opinion
network data must carefully choose their social network data—available seekers, leaders spread positive or negative information about a
data might not be suitable for targeted marketing purposes. At the same new product (Rogers, 1962, 2003). Recent studies show a low
time, the findings highlight the potential benefit of social network anal- correlation between the related but distinct concepts of opinion
ysis over the more traditional use of psychographic concepts. Finally, the leadership and social connectivity (Iyengar et al., 2011; Molitor,
empirical results offer new and surprising recommendations to market- Hinz, & Wegmann, 2011).
ing managers wanting to target consumers with the greatest influence, Psychographics also encapsulate consumers' independence with
challenging the common belief that targeting advisers is most effective. respect to judgments and desire to seek new product information.
2838 O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844

Two well-established constructs capture consumer innovativeness: their recording cannot reveal asymmetric friendships. Subsequent anal-
consumer independence of judgment making (CIJM) and consumer yses need to rely on less informative undirected data, even though the
novelty seeking (CNS) (Manning, Bearden, & Madden, 1995). People underlying social network structure actually might be directed.
differ in their level of reliance on others for information and assis-
tance in new product adoption decisions. Early adopters typically do 3.2. Determining centrality
not rely on others' opinions. In theory, high CIJM people take product
adoption risks without confirmatory information seeking from social Social network centrality relies on a global or local perspective
systems and they adopt new products early (Midgley & Dowling, (Bavelas, 1950). A globally central consumer maintains a position of
1978). In contrast, CNS represents consumers' motivation for ob- strategic significance in the network's overall structure; however, local
taining information about new products. Uncertainty surrounding a centrality applies to consumers with many social connections within
new product adoption decision diminishes as new information be- their immediate environment (Scott, 2000). This study applies mea-
comes available. High CNS people tend to adopt new products sooner sures of local centrality because the immediate environment of con-
than low CNS people (Hirschmann, 1980). sumers should be most influential for new product adoption decisions.
Depending on the network type (directed or undirected), calculating
3. Method network metrics relies on established methods from mathematical
graph theory (Barnes & Harary, 1983; Scott, 2000). To determine a
To measure social contagion in new product adoption behavior, this consumer's position in the local social network, this study applies the
study draws on concepts from social network analysis and traditional metric of degree-based centrality. Specifically, in undirected networks,
psychographic constructs. Interpersonal relationship data help explain the degree of consumer i indicates the consumer's number of social con-
the information flow through social networks and the social influence nections (see Fig. 1). To calculate the degree, let cij equal 1 when a con-
processes. Psychographic metrics capture consumers' attitudes toward nection exists between i and j. Then, the degree of i is as follows:
new product adoption and their self-assessment of their influence
over their peers. Previous research indicates that these two approaches degreei ¼ ∑ cij :
i≠j
yield significantly correlated results. Opinion leaders typically are high-
ly connected people; however, they also reflect different perspectives,
Directed networks indicate different numbers of incoming and
as some studies show a rather low correlation (e.g. Iyengar et al., 2011).
outgoing arrows. The number of incoming arrows to a point is the
This study also distinguishes between friendships and adviser–ad-
indegree. In turn, aji takes a value of 1 if an arrow moves from point
visee relationships. Friends tend to look out for each other in many
j to point i. Then, the indegree of i is as follows:
situations, whereas advisers support their advisees in specific areas
and with specific tasks. In the adviser–advisee relationship, A could
indegreei ¼ ∑ aji :
be an adviser for B for technology adoption, while B could be an ad- j≠i
viser for A for medicine.
For social network analysis, indegree reveals the number of people
3.1. Undirected and directed networks who considers a certain person their friend or adviser. In Fig. 1, B has
zero indegrees; A, D, and E each have one; and C has two. Incoming
The most common representation of social structures' formal arrows indicate appreciation, authority, or valuation as determined
properties is the sociogram (Fig. 1), a graphical diagram in which in- by other network members. A person with a high indegree tends to
dividuals appear as nodes (see A–E) and social relationships appear as emerge as an expert, adviser, or socio-metric opinion leader (Van
lines or arrows. Networks in which the relationship's direction is im- den Bulte & Wuyts, 2007; Wassermann & Faust, 1994).
portant are directed networks (Scott, 2000). Direction results from Analogously, the outdegree of point i is the total number of nodes
asymmetric relationships. For example, one person names the other to which i points:
as a friend or adviser, but not vice versa. In an advice network, an
arrow indicates “follows the advice of” and thus moves from the ad- outdegreei ¼ ∑ aij :
j≠i
visee (e.g., B) to the adviser (C). In contrast, in undirected networks
all relationships are symmetric: if A is C's friend then C is A's friend.
Some networks inherently are undirected (e.g., colleagues), and In Fig. 1, the outdegrees equal zero for nodes D and E, one for A
others are directed (e.g., child–parent). For friendship and advice net- and B, and two for C.
works, the distinction is unclear. Friendships should be symmetric; This study extends the general definition of degree by introducing
yet B might consider C a friend, even if C does not feel the same adopter-filtered degree, or the number of arrows connecting consum-
way about B. Similar asymmetry holds true for advice networks. To er i with others who already adopted. Let cij equal 1 if a connection
discover a social network's direction, the data must be available. For exists between points i and j, and let θj equal 1 if person j already
example, measuring friendships only after both parties agree to adopted (0 if not). Then, the adopter-filtered degree of i is as follows:

af:degreei ¼ ∑ cij θj :
i≠j
B D
Similarly, adopter-filtered indegree signifies the number of incom-
ing arrows from consumers who already adopted (i.e., the number of
advisees adopting a certain product). The adopter-filtered outdegree
measures the number of outgoing arrows. In an advice network, these
arrows signify the number of advisers who adopted the product. Let
aji take a value of 1 if a connection exists from point j to point i, and
let θj equal 1 if person j already adopted (and 0 if not). Then, the
adopter-filtered indegree and outdegree metrics are as follows:
A C E
af:indegreei ¼ ∑ aji θj and af:outdegreei ¼ ∑ aij θj :
Fig. 1. Sociogram. j≠i j≠i
O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844 2839

3.3. Controlling for individual psychographics Table 1


2 × 2 matrix of evaluated social networks.

The opinion leadership scale consists of six items related to opin- Friendship network Advice network
ion leadership (OL) and six items related to opinion seeking (OS)
Undirected network I II
(Flynn, Goldsmith, & Eastman, 1996; King & Summers, 1970). This Directed network III IV
study also operationalizes two aspects of innovativeness: CIJM and
CNS. These two reflexive constructs use scales that comprise six and
eight measurable items, respectively (Manning et al., 1995). undirected networks are reciprocal, the arrow is bidirectional regard-
less of whether B also named A. In the two directed networks (one
4. Empirical study friendship, one advice; networks III and IV in Table 1), uni- and bidi-
rectional relationships differ. If A names B as friend or adviser, an
4.1. Setting arrow from A to B enters the network. However, the arrow becomes
bidirectional only if B also names A. The directed‐networks' creation
This study focuses on social networks in secondary schools, fol- also requires consideration of purchase timing. If one consumer adopts
lowing Wassermann and Faust's (1994) assertion that the greatest before another, influence can only work in one direction. Applying this
share of social interaction occurs among students from the same constraint on the direction of social influence (see Section 3.2) in the
grade in the same school, due to everyday personal contact. Focusing subsequent analyses should improve network data quality (Hinz &
on students, the present study can pinpoint social contagion effects Spann, 2008).
with particular clarity and avoids unsuitable boundary specifications The students' social relationships, indicated by surveys, thus serve
that can harm empirical studies. However, this approach bears the as a basis for discerning undirected and directed networks of friend-
risk of overestimating the absolute social network impact if students ship and advice (2 × 2 matrix, four networks total; see Table 1). The
are more susceptible to social influences than the general adult pop- study analyzes these networks with the social network modeling pro-
ulation. In contrast, relative impacts should not be adversely affected. gram Pajek (Nooy, Mrvar, & Batagelj, 2005).

4.2. Questionnaire design


4.4. Data validity
A questionnaire collected information about students' social net-
work, adoption behavior toward specific products, psychographic Students in two secondary schools—one mixed gender and one all
metrics, and some demographics. The questionnaire includes four girls—took the survey during regular class time and under their
sections. First, students responded to psychometric questions about teachers' supervision, which resulted in high response rates, partici-
latent constructs, such as opinion leadership (OL) and opinion seek- pant attention, and accuracy. The schools' principals allowed distribu-
ing (OS), on seven-point Likert scales (1 = “strongly disagree,” 7 = tion of the survey to all students in the two highest grades and
“strongly agree”; see Flynn et al., 1996). The first section also contains provided lists of names that helped ensure both the completeness of
the latent CIJM and CNS constructs (Manning et al., 1995). Second, the the survey coverage and the proper coding of the network relation-
questionnaire elicited sociometric data with a fixed-choice method ships. The authors interviewed students who were not present for
(Wassermann & Faust, 1994). To distinguish different types of rela- the initial survey in the subsequent week and ultimately obtained re-
tionships in secondary school social networks, this study focuses on sponses from 342 of the 365 potential participants (94% response
friendship and advisory relationship networks. Participants named rate). The participants were 281 female and 53 male students, aged
up to eight fellow students they considered friends and eight fellow 15–20 years, with more than 90% in the age range of 17–19 years.
students they considered advisers with respect to consumer electron- The coding phase treated invalid responses (e.g., a claim of pur-
ics products. Third, items on adoption behavior toward Microsoft's chasing a PlayStation 3 in 2005, though this device became available
Xbox, Sony's PlayStation 3, Nintendo's Wii, and Apple's iPod followed for sale only in 2007) as missing data. Thus, 92% of all nominated
the interview guidelines of Coleman et al. (1966). One question elic- friends/advisers in the second survey section could be identified in
ited whether the respondent already adopted the product; if so, the the sample. Only eight percent of reported social network relation-
respondent indicated the date of purchase. If not, the respondent ships were treated as missing.
reported individual intention to buy that product in the next six An exploratory factor analysis assessed the psychometric proper-
months (seven-point scale, 1 = “very unlikely,” 7 = “very likely”; ties of the measures for the latent psychographic constructs: OL, OS,
Davis, 1989). Such self-predictions (behavioral expectations) are CIJM, and CNS. All scales are reliable. The mean scores for OL are sim-
highly accurate predictors of future behavior (Sheppard, Hartwick, ilar to those reported by the construct authors; mean scores for OS
& Warshaw, 1988; Warshaw & Davis, 1985). Fourth, participants pro- range slightly above previously reported values. For both constructs,
vided demographic information. the observed standard deviations are less than those that Flynn et
A study design pretest included 15 respondents who participated al. (1996) report, supporting the results. Mean scores for CIJM are
in personal, face-to-face interviews. These interviews aimed to reveal consistent with expected values (see Manning et al., 1995). Although
any comprehension problems with the questionnaire (Presser & Blair, the mean CNS scores lie below the expected values, they remain
1994) and helped identify and eliminate any ambiguities. within the range of reasonable results.
The Cronbach's alpha values (standardized items) are .73 for OL, .87
4.3. Construction of social networks for OS, .86 for CIJM, and .90 for CNS, all greater than the recommended
level of .7 to support scale reliability (Nunnally & Bernstein, 1994). As
The calculation of the network metrics relied on sociometric Churchill (1979) and Bagozzi and Phillips (1982) propose, this present
choice data (Section 3.2) and the friendships and advice relationships study uses the sum of the indicator scores as aggregated scales for
that students reported (Section 4.2). Table 1 shows the resulting four each construct (see also Flynn et al., 1996; Manning et al., 1995).
social networks. A post-hoc partial correlation procedure with a marker variable test-
The two undirected networks (one friendship, one advice; net- ed for common method bias. The smallest positive value in the matrix of
works I and II in Table 1) resulted if student A named student B as variable correlations functioned as a proxy for common method vari-
one of his or her top eight friends or advisers, in which case a bidirec- ance. The partial correlation adjustment reveals that zero-order correla-
tional arrow connects A and B by definition. Because relationships in tions that were statistically significant remain significant, so a common
2840 O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844

method bias is not a concern (Malhotra, Patil, & Kim, 2007; Podsakoff, Table 1) shows 63% reciprocal relationships, similar to the 70% order
MacKenzie, Lee, & Podsakoff, 2003). of magnitude that Kratzer and Lettl (2009) report among younger
children.
4.5. Characteristics of resulting social networks
4.6. Statistical analysis
The sociometric choice data enable social network construction
depicting students' relationships. For example, the directed-advice A repeated measures analysis of variance (ANOVA) includes inten-
network (network IV in Table 1) contains 334 nodes for the valid tion to adopt (ITA) as the dependent variable; network metrics as in-
data sets, with 873 directed-advice relationships depicted by arrows. dependent variables; and psychographic constructs, gender, and age
For readability, Fig. 2 only depicts a subnetwork with 190 nodes and as control variables. The students' grade is not a control variable, be-
501 arrows, based on the two-dimensional Fruchtermann–Reingold cause grade in school is strongly correlated with student age (Pearson
algorithm (Fruchterman & Reingold, 1991; Nooy et al., 2005), with ρ = .621, p b .001), and age should be more useful for comparisons
limited manual modifications. The colored nodes indicate students' with other studies. Adopters earned a maximum ITA score of seven;
grade level: blue for 13th grade, red for 12th grade. Few adviser refer- otherwise, self-reported intention to adopt provided the ITA value.
rals occur between grades, but many referrals exist within each grade. Another repeated measures ANOVA, with both treatment-varying
Of the 334 students, the self-reported consumer electronics product (ITA) and treatment-invariant (mixed effects) covariates, accounts
ownership measure indicates that nine own a Microsoft Xbox, nine own for the within-subject correlation of the four dependent variable
Sony PlayStation 3 consoles, 25 own a Nintendo Wii, and 97 own Apple scores for a student's intention to adopt the four products. A Wald
iPods. In addition, 213 students own none of the products, 105 own one, test analyzed the covariance distribution (assumed different for
and 16 own two or more. For all four products, the study asked the each product) for all ANOVA models.
non-adopters for their purchase intentions. Fig. 3 illustrates the collect-
ed data for the iPod advice network: red nodes represent iPod owners, 4.7. Results for undirected networks
yellow nodes refer to students with a high intention to adopt (score of 4
or higher on the seven-point Likert scale), and gray nodes refer to stu- The first analysis focuses on undirected networks (I and II, Table 1)
dents with a low intention to adopt (score of 3 or less). and thus deliberately disregards information on direction that is
Figs. 2 and 3 illustrate that adviser relationships are largely not re- available for this study but might not be available in many real-
ciprocal—if student A names student B as his or her adviser, only 31% world situations. The resulting friendship network is comparable to
of all cases does student B also name student A as an adviser. As a re- Facebook, where directional information on the relationships is missing.
sult, only 271 bidirectional relationships occur in the total advice net- Similarly, the undirected-advice network is comparable to LinkedIn. Ex-
work. In contrast, the directed-friendship network (network III in tant literature offers no predictions on whether undirected-friendship

Fig. 2. Sociogram for social network of mutual advice.


O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844 2841

Fig. 3. Sociogram for mutual advice with intentions to adopt the iPod.

or undirected-advice networks are better suited to identify influential for adopter-filtered outdegree and adopter-filtered indegree are .173
consumers. The hype and hope around social network platforms and .537, respectively. Results from the friendship network are not as
such as Facebook suggest, however, that practitioners widely believe clear: af.outdegree is barely significant, and af.indegree shows no signif-
that friendship data help them target influential consumers. Study re- icance. For social network sites that capture friendships (e.g., Facebook),
sults show the number of adopters in a personal social network these results indicate that adding directional information (e.g., rejected
measured by the adopter-filtered degree (af.degree, see Section 3.2) or unanswered friendship requests) does not make the networks as
significantly and positively impact the ITA. The estimated coefficient valuable as directed-advice networks for targeting consumers. In con-
for the undirected-advice network is .325, approximately five times trast, undirected-advice networks such as LinkedIn could benefit from
greater than the value in a corresponding model based on friendship net- capturing directional information. Google+ captures directional infor-
works. Furthermore, the af.degree's significance clearly increases in the mation and can model various topical subnetworks. This social net-
advice network model as well as the improved log-likelihoods of the working platform should provide firms richer information to identify
advice model. Table 2 summarizes the results. These findings suggest and to target influential consumers.
undirected-advice networks (e.g., LinkedIn) outperform undirected- Coefficient estimates and significance levels only marginally
friendship networks (e.g., Facebook) with respect to explanatory change when controlling for psychographic constructs. These findings
power. These findings are somewhat surprising. The evidence suggests confirm that ego-centric, psychographic indicators for social connect-
that researchers, practitioners, and investors should use social network edness do not capture all information in a social network. Despite a
data from undirected-friendship networks only with caution to identify rather high correlation between OS and CIJM, removing one measure
and target influential consumers. If possible, they should rely on advice from the model does not change the estimates substantially, so the
networks due to their greater explanatory power. collinearity of these two variables is not an issue.
Adding psychographic constructs changes the estimates of the This analysis also shows that the number of adopters a person per-
remaining variables only slightly, and only OL and CNS are significant. ceives as advisers for new product‐related questions (adopter-filtered
Thus, the explanatory power from structural data goes beyond the outdegree) positively influences the intention to adopt the new product.
traditional ego-centric psychographic constructs used in this study. This finding confirms conventional wisdom: if an adviser has already
bought a specific product, he or she influences his or her advisees,
4.8. Results for directed networks whose likelihood of buying that product increases. One additional adopt-
er among a person's advisers increases that person's intention to adopt by
The second analysis includes directed social network information approximately three percent (if the seven-point Likert scale represents
(see Table 3). The results correspond to the analysis of social network 100% of a person's adoption probability), even after controlling for indi-
III and IV in Table 1. For the advice network, the significant coefficients vidual psychographic data.
2842 O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844

Table 2
Mixed-effects models for ITA based on undirected networks.

Variable Undirected-friendship network Undirected-advice network

Basic + Psychographics Basic + Psychographics

Coeff. (std. error) df Coeff. (std. error) df Coeff. (std. error) df Coeff. (std. error) df

Intercept 2.153 (1.34) 326 .846 (1.44) 309 1.674 (1.33) 323 .359 (1.43) 308
Product 1 (Xbox) −2.296 *** (.17) 1127 −2.301 *** (.18) 1,104 −2.058 *** (.15) 1057 −2.063 *** (.16) 1,033
Product 2 (PS3) −2.021 *** (.18) 1113 −2.022 *** (.18) 1,110 −1.770 *** (.15) 1063 −1.772 *** (.16) 1,039
Product 3 (Wii) −1.714 *** (.16) 1101 −1.735 *** (.17) 1,079 −1.512 *** (.15) 1036 −1.526 *** (.15) 1,014
Product 4 (iPod) 0 a 0 a 0 a 0 a
Age .088 (.07) 316 .072 (.07) 303 .100 (.07) 316 .084 (.07) 302
Gender −.163 (.16) 316 .089 (.16) 304 −.145 (.16) 315 .116 (.16) 303
af.degree .070 * (.03) 1269 .070 * (.03) 1,230 .325 *** (.05) 1266 .322 *** (.05) 1,230
OLc – .025 * (.01) 302 – .023 * (.01) 302
OS – .000 (.01) 304 – −.001 (.01) 302
CIJM – .012 (.01) 303 – .015 (.01) 304
CNS — .023 *** (.01) 303 — .023 *** (.01) 302
Log-likelihood −2570.443 −2503.232 −2553.064 −2487.073

a: Parameter set to zero because it is redundant. Significance levels: ***.001 **.01 *.05.

The results also highlight a less intuitive effect in the reverse di- considerations. Surprisingly, the results reveal that the mechanism
rection. Pressure arises from structural equivalence: The number of of structural equivalence in particular and also status considerations
advisees who adopt the new product influences the adviser's own in- can drive adoption. Cohesion also affects behavior, but at a much
tention to adopt. If an advisee adopts, but the adviser has not, social smaller magnitude. Social cohesion's different aspects become visible
pressure increases the adviser's level of intention to adopt. This ob- only with directed information in adviser–advisee networks. Mere
servation suggests advisers attempt to defend their social position friendship information is insufficient for precise and effective
within the immediate social network (Burt, 1987). This effect's mag- targeting. This finding parallels recent conjectures by Van den Bulte
nitude is surprising: One more adopter among advisees increases the (2009) and Hubermann, Romero, and Wu (2009): Evaluating inter-
adviser's intention to adopt by .52 on the seven-point Likert scale— personal relationships from the wrong perspective, or not filtering
that is, social pressure increases adoption probability by 7%. out irrelevant links, produces dubious results.
Controlling for classical concepts of opinion leadership (Flynn et al.,
4.9. Influence of age, gender, and interactions 1994) and innovativeness (Manning et al., 1995) avoids confounding
the social networks' impact on adoption decisions with individual psy-
In all the models, neither age nor gender indicates statistical sig- chographics. The study shows that ego-centric psychographic metrics
nificance (p > .1). This study therefore does not confirm the intuition can enhance, but are by no means substitutes for, the results derived
that age or gender differences alter adoption decisions (e.g., Gefen & from analyzing consumers' social network structures. The effect of so-
Straub, 1997). Interaction effects in the various models (e.g., between cial contagion through social networks goes beyond the person's psy-
OL and degree metrics) also are not statistically significant and they chographic characteristics and further illustrates that economic
reduce models' log-likelihood. decision making is embedded in his or her environment (Granovetter,
1985).
5. General discussion
5.2. Managerial implications
5.1. Research contribution
Social network platforms' success makes using data on friendships
This study confirms that social contagion significantly drives new to explain and ultimately influence economic behavior appealing to
product adoption behavior (Coleman et al., 1966). This study's unique firms and researchers. Firms hope that extensive social networks
data allow distinguishing between the effect of cohesion and status prove capable of identifying consumers with the greatest influence

Table 3
Mixed effects models for ITA based on directed networks.

Variable Directed-friendship network Directed-advice network

Basic + Psychographics Basic + Psychographics

Coeff. (std. error) Df Coeff. (std. error) df Coeff. (std. error) df Coeff. (std. error) df

Intercept 2.160 (1.34) 330 .892 (1.44) 314 1.664 (1.32) 322 .384 (1.43) 308
Product 1 (Xbox) −2.286 *** (.17) 1141 −2.273 *** (.18) 1118 −2.050 *** (.15) 1,057 −2.053 *** (.16) 1033
Product 2 (PS3) −1.993 *** (.18) 1147 −1.972 *** (.18) 1124 −1.779 *** (.15) 1,064 −1.777 *** (.16) 1039
Product 3 (Wii) −1.727 *** (.16) 1115 −1.729 *** (.17) 1093 −1.539 *** (.15) 1037 −1.536 *** (.15) 1015
Product 4 (iPod) 0 a 0 a 0 a 0 a
Age .089 (.07) 321 .075 (.07) 308 .099 (.07) 315 .083 (.07) 302
Gender −.188 (.16) 321 .087 (.16) 309 −.128 (.16) 315 .121 (.16) 302
af.outdegree .102 (.08) 1284 .139 ° (.08) 1250 .173 * (.08) 1268 .187 * (.08) 1230
af.indegree .043 (.06) 1286 .015 (.07) 1,243 .537 *** (.10) 1263 .521 *** (.10) 1226
OL – .023 * (.01) 308 – .022 * (.01) 302
OS – −.004 (.01) 309 – .000 (.01) 302
CIJM – .012 (.01) 308 – .014 (.01) 305
CNS – .025 *** (.01) 309 – .022 *** (.01) 302
Log-likelihood −2612.822 −2544.278 −2550.611 −2485.272

a: Parameter set to zero because it is redundant. Significance levels: ***.001 **.01 *.05 °.1.
O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844 2843

on others' adoption behavior so that they can target them. Re- Fourth, the study does not control for mass media impacts on
searchers hope to find explanatory power in social network analysis, purchase decisions and thus covers only the second aspect of the hypoth-
beyond that of traditional models with ego-centric psychographic esized two-step flow process of new product adoption (Lazarsfeld,
metrics. This study offers new and somewhat surprising findings re- Berelson, & Gaudet, 1944). Because the study includes an analysis of
garding the suitability of social networks for explaining and ultimate- adoption behavior data for four different consumer electronics products
ly influencing consumers' new product adoption behavior. and indicates significance in the models, the mass media impact on
First, this research compares the explanatory power of four types of adoption decisions likely is either irrelevant or equal for all products. Fur-
networks (Table 1), contrasting undirected networks, in which relation- ther research should investigate this conclusion and test the robustness
ships among consumers are reciprocal, with directed networks, in of the findings while controlling for mass media's impact.
which roles are clearly separate, and contrasting advice networks, in
which relationships revolve around a specific topic, with friendship net- Acknowledgments
works, with a more general orientation. Advice networks' explanatory
power clearly exceeds pure friendship networks. This result issues a The authors acknowledge and are grateful for the support by
warning to firms and researchers who analyze data currently available Ghenet Abraha and the financial support by the E‐Finance Lab,
in social network platforms primarily on the basis of relationships be- Frankfurt. Comments by Ju‐Young Kim (Goethe‐University Frankfurt),
tween friends. In addition, directed-friendship networks are not superi- Gal Oestreicher‐Singer (Tel Aviv University), Christophe van den
or to undirected-friendship networks in explanatory power. Even if a Bulte (University of Pennsylvania), Bernd Skiera (Goethe‐University
network such as Facebook records direction information, the usefulness of Frankfurt), and Martin Spann (Ludwig‐Maximilians‐University
for targeting purposes remains limited. In contrast, advice networks Munich) to earlier drafts were helpful in revising this study. Two anon-
could benefit substantially from directed information. Google + offers ymous referees and the associate editor Drew Martin also offered excel-
an example of how to combine an advice network and directional infor- lent input. The authors alone are responsible for all limitations and
mation within a social network platform. errors that may relate to the study and the paper.
Second, this study reveals that the influence of advisees who
adopt a new product on their advisers is more than three times great- References
er than the influence of advisers who adopt on their advisees. These
Abrahamson, E., & Rosenkopf, L. (1997). Social network effects on the extent of innova-
results are robust to controlling for the individual psychographic con-
tion diffusion: A computer simulation. Organization Science, 8(3), 289–309.
structs of opinion leadership and innovativeness. Advisees' early Algesheimer, R., Borle, S., Dholakia, U. M., & Singh, S. S. (2010). The impact of customer
adoption behavior pressures non-adopting advisers to adopt to pre- community participation on customer behaviors: An empirical investigation. Mar-
keting Science, 29(4), 756–769.
serve their social status. Firms should consider more sophisticated
Bagozzi, R. P., & Phillips, L. W. (1982). Representing and testing organizational theories:
targeting strategies that incorporate targeting advisees, not just A holistic construal. Administrative Science Quarterly, 27(3), 459–489.
advisers. Bampo, M., Ewing, M. T., Mather, D. R., Stewart, D., & Wallace, M. (September). The
effects of the social structure of digital networks on viral marketing performance.
Information Systems Research, 19, 273–290.
5.3. Limitations and directions for further research Barnes, J. A., & Harary, F. (1983). Graph theory in network analysis. Social Networks,
5(2), 235–244.
Regarding the magnitude of the social network's impact on adoption Bavelas, A. (1950). Communication patterns in task- oriented groups. The Journal of the
Acoustical Society of America, 22(6), 725–730.
behavior, four limitations arise. First, conducting a survey with Borgatti, S. P., & Cross, R. (2003). A relational view of information seeking and learning
near-adult secondary school students creates a risk of biased estimates in social networks. Management Science, 49(4), 432–445.
of the impact of the social network on adoption decisions. Social net- Bowler, W. M., Dahlstrom, R., Seevers, M. T., & Skinner, S. J. (2011). The ties that buy:
the role of interfirm social contagion across customer accounts. Journal of Personal
work density and mutual influence presumably decrease after students Selling & Sales Management, 31(1), 7–20.
leave school. This approach might over-estimate the absolute effect Burt, R. S. (1987). Social contagion and innovation: Cohesion versus structural equivalence.
magnitude in the analyses, but the relative effects within (i.e., influence The American Journal of Sociology, 92(6), 1287–1335.
Churchill, G. A. (1979). A paradigm for developing better measures of marketing con-
of adopters on advisers compared with adopters on advisees) and structs. Journal of Marketing Research, 16(1), 64–73.
across (i.e., explanatory power of advice vs. friendship or directed vs. Coleman, J. S., Katz, E., & Menzel, H. (1966). Medical innovation: A diffusion study.
undirected networks) the network types studied should hold. An up- Indianapolis: Bobbs- Merrill Co.
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of
ward bias strengthens the findings because the explanatory power of
information technology. MIS Quarterly, 13(3), 319–340.
friendship networks for non-student populations should be even Festinger, L., Schachter, S., & Back, K. W. (1950). Social pressures in informal groups: A
lower than reported results. study of human factors in housing. Stanford California: Stanford University Press.
Second, stated choices and preferences in a questionnaire feature Fischer, C. (1978). Urban-to-rural diffusion of opinions in contemporary America. The
American Journal of Sociology, 84(1), 151–159.
some well-known drawbacks compared with revealed behavior Flynn, L. R., Goldsmith, R. E., & Eastman, J. K. (1994). The king and summers opinion
(Swait & Andrews, 2003). For this study, the questionnaire also offers leadership scale: Revision and refinement. Journal of Business Research, 31(1),
important advantages. For example, the questionnaire extracts reliable 55–64.
Flynn, L. R., Goldsmith, R. E., & Eastman, J. K. (1996). Opinion leaders and opinion
information on two types of directed networks (friendship and advice) seekers: Two new measurement scales. Journal of the Academy of Marketing Research,
with well-defined boundaries, as well as indications of opinion leader- 24(2), 137–147.
ship, innovativeness, and key demographics. Other data usually do not Fruchterman, T. M. J., & Reingold, E. M. (1991). Graph drawing by force-directed place-
ment. Software: Practice and Experience, 21(11), 1129–1164.
support easy comparisons of two types of networks for the same popu- Gefen, D., & Straub, D. W. (1997). Gender differences in the perception and use of
lation, especially if the study requires directed information on both e-mail: An extension to the technology acceptance model. MIS Quarterly, 21(4),
network types. In addition, evaluating the social network data's explan- 389–400.
Godes, D., & Mayzlin, D. (2009). Firm-created word-of-mouth communication: Evidence
atory power using psychographic concepts requires additional informa- from a field test. Marketing Science, 28(4), 721–739.
tion on opinion leadership and innovativeness. Observed information Goldenberg, J., Han, S., Lehmann, D. R., & Hong, J. W. (2009). The role of hubs in the
unlikely provides all the data required for this study. adoption process. Journal of Marketing, 73(2), 1–13.
Granovetter, M. (1985). Economic action and social structure: The problem of
Third, the sample is biased toward female respondents. Although
embeddedness. The American Journal of Sociology, 91(3), 481–510.
previous work indicates that women socially communicate and be- Hinz, O., Skiera, B., Barrot, C., & Becker, J. U. (2011). Seeding strategies for viral marketing:
have differently regarding new technology, study results suggest gen- An empirical comparison. Journal of Marketing, 75(6), 55–71.
der does not influence product adoption intent. Recent studies show a Hinz, O., & Spann, M. (2008). The impact of information diffusion on bidding behavior
in secret reserve price auctions. Information Systems Research, 19(3), 351–368.
smaller gender divide with respect to the products studied (Lewis, Hirschmann, E. C. (1980). Innovativeness, novelty seeking, and consumer creativity.
2007; Solutions Research Group Consultants, 2010). Journal of Consumer Research, 7(3), 283–295.
2844 O. Hinz et al. / Journal of Business Research 67 (2014) 2836–2844

Hubermann, B. A., Romero, D. M., & Wu, F. (2009). Social networks that matter: Twitter Presser, S., & Blair, J. (1994). Survey pretesting: Do different methods produce different
under the microscope. Social Computing Laboratory, HP Labs. results? Sociological Methodology, 24, 73–104.
Iyengar, R., Van den Bulte, C., & Valente, T. W. (2011). Opinion leadership and social Rogers, E. M. (2003). Diffusion of innovations. New York: Free Press.
contagion in new product diffusion. Marketing Science, 30(2), 195–212. Rogers, E. M., & Shoemaker, F. F. (1971). Communication of innovations: A cross-cultural
Katz, E., & Lazarsfeld, P. F. (1955). Personal influence: The part played by people in the approach (2nd ed.). New York: Free Press.
flow of mass communications. New York: Free Press. Sailer, L. D. (1978). Structural equivalence: Meaning and definition, computation and
King, C. W., & Summers, J. O. (1970). Overlap of opinion leadership across consumer application. Social Networks, 1(1), 73–90.
product categories. Journal of Marketing Research, 7(1), 43–50. Schmitt, P., Skiera, B., & Van den Bulte, C. (2011). Referral programs and customer
Kratzer, J., & Lettl, C. (2009). Distinctive roles of lead users and opinion leaders value. Journal of Marketing, 75(1), 46–59.
in the social networks of schoolchildren. Journal of Consumer Research, 36(4), Scott, J. (2000). Social network analysis: A handbook. London: Sage.
646–659. Sheppard, B. H., Hartwick, J., & Warshaw, P. R. (1988). The theory of reasoned action: A
Lazarsfeld, P. F., Berelson, B., & Gaudet, H. (1944). The people's choice: How the voter meta-analysis of past research with recommendations for modifications and future
makes up his mind in a presidential campaign. New York: Duell Sloan and Pearce. research. Journal of Consumer Research, 15(3), 325–343.
Lewis, L. (2007). Nintendo's women gamers could transform market. Times Online. Solutions Research Group Consultants (2010). Who wants a tablet? Everybody.
http://business.timesonline.co.uk/tol/business/markets/japan/article2629649.ece Swait, J., & Andrews, R. L. (2003). Enriching scanner panel models with choice experiments.
Accessed August 2009. Marketing Science, 22(4), 442–460.
Libai, B., Bolton, R., Bügel, M. S., de Ruyter, K., Götz, O., Risselada, H., et al. (2010). Trusov, M., Bucklin, R. E., & Pauwels, K. (2009). Effects of word-of-mouth versus tradi-
Customer-to-customer interactions: Broadening the scope of word of mouth re- tional marketing: Findings from an Internet social networking site. Journal of Mar-
search. Journal of Service Research, 13(3), 267–282. keting, 73(5), 90–102.
Mahajan, V., & Muller, E. (1979). Innovation diffusion and new product growth models Valente, T. W. (1996). Social network thresholds in the diffusion of innovations. Social
in marketing. Journal of Marketing, 43(4), 55–68. Networks, 18(1), 69–89.
Malhotra, N. K., Patil, A., & Kim, S. S. (2007). BIAS breakdown. Marketing Research, Van den Bulte, C. (2009). Opportunities and challenges in studying consumer net-
19(1), 24–29. works. In S. Wuyts, M. G. Dekimpe, E. Gijsbrechts, & R. Pieters (Eds.), The connected
Manning, K. C., Bearden, W. O., & Madden, T. J. (1995). Consumer innovativeness and customer. London: Routledge.
the adoption process. Journal of Consulting Psychology, 4(4), 329–345. Van den Bulte, C., & Stremersch, S. (2004). Social contagion and income heterogeneity
Midgley, D. F., & Dowling, G. R. (1978). Innovativeness: The concept and its measure- in new product diffusion: A meta-analytic test. Marketing Science, 23(4), 530–544.
ment. Journal of Consumer Research, 4(4), 229–242. Van den Bulte, C., & Wuyts, S. (2007). Social networks and marketing. Cambridge, MA:
Molitor, D., Hinz, O., & Wegmann, S. (2011). The interplay between psychometric and Marketing Science Institute.
sociometric data and the willingness to adopt new products. Zeitschrift für Villanueva, J., Yoo, S., & Hanssens, D. M. (2008). The impact of marketing-induced
Betriebswirtschaft (ZfB), 81(1), 29–59. versus word-of-mouth customer acquisition on customer equity growth. Journal
Nooy, Wd, Mrvar, A., & Batagelj, V. (2005). Exploratory social network analysis with of Marketing Research, 45(1), 48–59.
Pajek. New York: Cambridge University Press. Warshaw, P. R., & Davis, F. D. (1985). Disentangling behavioral intention and behavioral
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3rd ed.). New York: expectation. Journal of Experimental Social Psychology, 21(3), 213–228.
McGraw-Hill. Wassermann, S., & Faust, K. (1994). Social network analysis: Methods and applications.
Podsakoff, P. M., MacKenzie, S. B., Lee, J. -Y., & Podsakoff, N. P. (2003). Common method Cambridge, New York: Cambridge University Press.
biases in behavioral research: A critical review of the literature and recommended Watts, D. J., & Dodds, P. S. (2007). Influentials, networks, and public opinion formation.
remedies. The Journal of Applied Psychology, 88(5), 879–903. Journal of Consumer Research, 34(4), 441–458.