Is Investing in Social Media Really Worth It? How Brand Actions and User Actions Influence Brand Value

Abstract

A lthough previous studies have documented a positive link between traditional media and brand performance, how social media is related to brand value has not yet been comprehensively explored. We propose a conceptual model to address this research gap, collecting a unique data set that captures information on user and brand actions on three social media platforms (Facebook, Twitter, and YouTube), word-of-mouth, and brand value for 87 brands in 17 industries. We empirically test our model with partial least squares path modeling (PLS-PM). First, we test the direct effects and find that user actions on YouTube and brand actions on Facebook have a positive influence on brand value. Second, we enrich our model by including word-of-mouth as a mediator, finding that the effect of social media goes above and beyond pure word-of-mouth spread. We test for alternative models, by first accounting for sample heterogeneity and second by including brand strength as a control variable, finding that the main model results' are indeed robust. Our study demonstrates that making use of social media positively relates to brand value, as well as validates a set of objective metrics to measure social media actions, thus advancing knowledge on social media marketing for both academics and practitioners.

Full text

SERVICE SCIENCE
Vol. 8, No. 2, June 2016, pp. 152–168
ISSN 2164-3962 (print) ISSN 2164-3970 (online) http://dx.doi.org/10.1287/serv.2016.0143
© 2016 INFORMS
Is Investing in Social Media Really Worth It? How
Brand Actions and User Actions Influence Brand Value
Anatoli Colicev
Nazarbayev University Graduate School of Business, Astana, Republic of Kazakhstan 010000, anatoli.colicev@nu.edu.kz
Peter O’Connor, Vincenzo Esposito Vinzi
Information Systems, Decision Sciences, and Statistics, ESSEC Business School,95201 Cergy-Pontoise Cedex, France
{oconnor@essec.edu,vinzi@essec.edu}
Although previous studies have documented a positive link between traditional media and brand performance, how social
media is related to brand value has not yet been comprehensively explored. We propose a conceptual model to address
this research gap, collecting a unique data set that captures information on user and brand actions on three social media
platforms (Facebook, Twitter, and YouTube), word-of-mouth, and brand value for 87 brands in 17 industries. We empirically
test our model with partial least squares path modeling (PLS-PM). First, we test the direct effects and find that user actions
on YouTube and brand actions on Facebook have a positive influence on brand value. Second, we enrich our model by
including word-of-mouth as a mediator, finding that the effect of social media goes above and beyond pure word-of-mouth
spread. We test for alternative models, by first accounting for sample heterogeneity and second by including brand strength as
a control variable, finding that the main model results’ are indeed robust. Our study demonstrates that making use of social
media positively relates to brand value, as well as validates a set of objective metrics to measure social media actions, thus
advancing knowledge on social media marketing for both academics and practitioners.
Keywords: social media; branding; brand value; partial least squares (PLS); return on investment
History: Received December 20, 2014; Received in revised form March 10, 2016; Accepted March 16, 2016. Published
online XXX, XX, XXXX.
Introduction
Few would deny the rapid evolution of social media. Relatively unknown 10 years ago, it now takes center stage
in many consumers’ lives, changing the nature of marketing communications. A recent report by Nielsen (2012)
shows that while 92% of consumers trust recommendations from people they know (for example Facebook
friends) and 70% trust reviews posted online (including anonymous comments on online review sites), only 47%
trust conventional advertising on television, newspapers, or radio. As consumers increasingly turn away from
traditional media (Mangold and Faulds 2009), establishing and effectively leveraging a social media presence
has become essential. Indeed, most companies have started to build their digital assets. Of the Fortune 500
companies, 73% have a Twitter account, 66% have a Facebook account, and 62% have a YouTube channel
(Heggestuen and Danova 2013). Academics have also acknowledged the rapid growth of social media, calling
for further research on social media accountability (see Kumar 2015) and stressed the latter’s priority for future
studies (Marketing Science Institute 2014). One of the key challenges is justifying the significant investment
needed to meaningfully engage in social media marketing (Deans 2011), with continuous requests for evidence
that demonstrates a positive return on social media spending (Weinberg and Pehlivan 2011).
While some attempts have been made to relate social media usage to performance indicators such as share-
holder value (see, e.g., Luo et al. 2013a), to date no studies have addressed how social media is related to
another key asset of the firm—brand value. Brands typically possess both tangible and intangible value. For
example, Interbrand values Coca Cola at US$77 billion, much higher than its market capitalization, a difference
attributed to the value of its multiple brands. Accordingly, this study develops a framework to assess the effects
of social media actions on brand value, thus addressing the current gap in the literature. Although it has yet
to be tested empirically, such an effect is anecdotally perceived to be positive for several reasons. First, social
media helps reduce brand-related information asymmetry by constituting a credible source of information for
consumers (Tirunillai and Tellis 2012). In addition, brands that successfully leverage social media typically
maintain extensive user networks and are thus able to benefit from electronic word-of-mouth dissemination
to large, essentially opt-in, audiences (Peters et al. 2013). Brands can also enhance customer expectations by
co-developing new product ideas with the community, as well as involving the latter in critical product-related
decisions (e.g., cocreation) (Schau et al. 2009). As a result, there is much evidence to suggest that engaging in
social media in a meaningful way would have a positive effect on brand value.
152

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 153
To test the above proposition we collect data from the three largest social media platforms (Facebook, Twitter,
and YouTube) as proxies for brand actions and user actions on social media. In common with other studies on
brand value (see, e.g., Torres et al. 2012), data from both Interbrand and Brand Finance is used as a quantitative
estimate of brand value, and we test our model using partial least squares path modeling (PLS-PM) with the
objective of establishing if, and which, social media metrics can be related to brand value.
Our study advances the literature by highlighting the relationship between social media actions and brand
value, thus offering implications for both theory and practice. By highlighting the role of social media in brand
value creation, we add to the collective understanding of the ways in which social media benefit businesses.
We find that social media actions are related to brand value both directly and indirectly (through the mediation
of word-of-mouth). In addition we provide empirical support for the growing importance of user actions on
social media, which we find to significantly relate to increases in brand value. While it is true that user actions
are not under the direct control of brands, we postulate that they can be influenced by brand actions, making
the role of social media marketing even more important. We find that word-of-mouth partially mediates these
relationships. Thus social media activities have a double role in determining brand value: they help disseminate
word-of-mouth that in turn leads to higher brand value, and they are still significantly related to brand value
even after accounting for the effects of word-of-mouth. This implies that social media activities are not simply
associated with information spread (WOM) but also constitute a valuable asset for increasing brand value.
Conceptual Background and Hypothesis Development
The Impact of Media on Brand Value
One of the main challenges in justifying marketing expenditures is identifying appropriate performance indicators
(Srinivasan and Hanssens 2009). Our study specifically examines how social media actions are associated with
brand value and thus directly addresses this highly topical issue.
It is well established that brands perform a valuable set of functions for both consumers and firms, and are
acknowledged to have substantial tangible and intangible value (Keller and Lehmann 2006). One of the most
commonly used approaches to estimating brand value is the Interbrand brand valuation method, a measure that
has already been widely used in academic research (see, e.g., Motameni and Shahrokhi 1998, Torres et al.
2012). Furthermore, Barth et al. (1998) found Interbrand’s valuations to be relevant and sufficiently reliable for
financial reporting statements.
Previous studies have already established that traditional advertising creates competitive advantage for a brand
(Mela et al. 1998), and has a direct effect on brand value (Simon and Sullivan 1993). However, as discussed
above, consumers are increasingly turning away from traditional advertising and placing increased trust in
social media when making purchasing decisions (Mangold and Faulds 2009). Brands that properly manage their
social media presences thus have an opportunity to enhance their reputation, brand awareness, and brand name
(Murdough 2009) as well as ultimately influence consumer purchase intentions (Kim and Ko 2012).
Despite this, most brands face challenges in calculating their return on investment on social media spend
(Hoffman and Fodor 2010). In particular, the vagueness associated with budget allocation, as well as the lack of
easily quantifiable return, leads to ambiguity and brings into question the validity of such calculations (Weinberg
and Pehlivan 2011). In addition, the resulting effect is often not immediately apparent, as it takes time for the
word-of-mouth to spread or buzz to develop, making it difficult to quantifiably justify the investment (Hoffman
and Fodor 2010).
A limited amount of literature has attempted to address this issue. Luo et al. (2013b) investigated nine
firms from the hardware and software industries, showing that buzz and Internet traffic drive shareholder value.
Similarly, Chen et al. (2012) found that online reviews can impact purchase intention. However, most of these
studies are limited either to a specific medium (e.g., online reviews on Amazon, or consumer buzz on the
Internet) or to a small sample composed of similar brands within a single industry. Thus, there is currently no
comprehensive and commonly accepted approach toward assessing the effect of using social media on brand
value.
To address this issue, we use a similar approach to Stephen and Galak (2012) by separating social media
actions in two distinct categories (brand actions and user actions), and relating these two activities to changes
in brand value. We use activity on the three largest social media platforms (Facebook, Twitter, and YouTube) as
our proxies for brand and user actions on social media. For example, brands can engage in social media actions
by posting new content on Facebook (videos, photos, status updates), tweeting on Twitter, or posting a video on
YouTube. Such brand actions not only potentially are related to brand value, but can also drive subsequent user

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
154 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Table 1. Measures and Data Sources
Construct Metrics Description Source
Brand actions on
Facebook
Number of photos posted
Number of status updates
Number of videos posted
Number of miscellaneous posted
Brand actions on Facebook
measured by the number
of brand posts that come
in different forms
Proprietary data
source
Brand actions on
Twitter
Number of tweets Brand actions on Twitter
measured by the number
of tweets
Proprietary data
source
User actions on
Facebook
Number of likes on brands’ page
Number of likes/brands’ posts
Number of comments/brands’ posts
Number of shares/brands’ posts
Number of people talking about this
(PTAT)
User actions on Facebook
measured by the number
of fans, PTAT, and
interactions on brands’
posts
Proprietary data
source
User actions on
Twitter
Number of followers
Number of retweets
User actions on Twitter
measured by the number
of brand followers and
retweets on brands’ tweets
Proprietary data
source
User actions on
YouTube
Number of subscribers
Number of video views
User actions on YouTube
measured by the number
of subscribers and video
views on brands’ channel
Proprietary data
source
Brand value Brand value Interbrand
Brand value Brand Finance
Brand values computed by
Interbrand and Brand
Finance
Interbrand and Brand
Finance brand value
databases
Note. All variables are operationalized as a difference between December 2013 and December 2012 value.
actions. In contrast, users actions can include, for example, liking the brand or its content, or posting a comment
on the brand’s posts on Facebook, following the brand or retweeting the brand’s content on Twitter, or watching
videos on YouTube. We therefore define our conceptual constructs as actions taken on social media channels
either by brands or users. We summarize and explain each of the constructs used in this study in Table 1.
Social Media Actions
Brand and User Actions on Facebook. Brands can use their Facebook page for a variety of purposes,
including disseminating information, interacting with their consumer base, and/or building a sense of community.
For example, Goh et al. (2013) demonstrate that marketers should be active on their Facebook brand page and
promote interactions with their content so as to generate favorable purchasing behavior. Responding promptly to
customer remarks helps create a positive image and build trust, thereby increasing the perceived quality and value
of the brand (Bruhn et al. 2012). Similarly, McCann (2013) suggests that responding to consumer complaints
through the Facebook brand page can make consumers feel more valued and encourage them to subsequently
recommend the brand. As such, a well-managed social media presence enables a high level of transparency,
fosters cooperation and trust, and potentially, positively impacts purchasing behavior (Hoffman and Fodor 2010),
and consequently brand value.
We operationalize the brand actions on Facebook construct as the number of brand posts on their own
Facebook page. Moreover, as people are most likely to engage with content on social media that contains pictures
(44%) and videos (37%) (Performics 2012), we split brand actions into status updates, video and photo sharing,
and other forms of posts to allow for different weights to be used in estimating the brand actions on Facebook
construct.
In addition, one of the goals of brand posts is typically to generate user actions through liking,commenting,
and sharing. We postulate that the brand actions on Facebook would be positively related to user actions on
Facebook, reflected in the number of likes on the brand page, the likes/comments/shares on brand posts, as well
as in the people talking about this metric (PTAT). A like is a sign that someone is interested in a brand and
indicates a willingness to engage. Users can not only like a brand page but also like, comment on, or share
brand posts. User actions in response to brand activity are shown not only on that user’s individual wall, but

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 155
also selectively disseminated toward the news feeds of their friend network. Thus, encouraging users to interact
is interesting as it spreads the brand’s message more widely. Last, the PTAT metric measures users’ voluntary
engagement in the form of storytelling about a brand. According to Mazin (2011) 41% of Facebook users
regularly share stories about brands, making PTAT an important metric for measuring user activity, as content
creators typically exhibit the highest level of loyalty (Turri et al. 2013).
As discussed above, user actions on Facebook form a reliable and credible source of brand-related information
for other users, thereby potentially reducing brand-related informational asymmetry (Luo et al. 2013a). Users
who actively engage with brand pages have also been found to be more loyal, leading to repeated purchases
and brand advocacy (Yadav et al. 2013). However, it should be noted that user actions can be either positive
or negative. In the latter case user actions on Facebook could potentially influence brand value negatively
based on unfavorable sentiment (see Yu et al. 2013 for a recent review and application of sentiment analysis).
Overall there is disagreement about how the valence of social media (and electronic word-of-mouth (eWOM)
in general) affects firm performance. Some researchers claim that just the volume (number) of online reviews
predicts product sales, irrespective of their valence (e.g., Duan et al. 2008, Ho-Dac et al. 2013). Others claim
that the main predictor is not the volume but rather its valence (e.g., Chintagunta et al. 2010, Dellarocas 2006),
variance (Sun 2012), or specific content (e.g., Onishi and Manchanda 2012).
More specifically, evidence on the impact of negative content is inconclusive (Babi´c et al. 2015). On the
one hand, negative sentiment should have a negative effect on consumers and some studies do indeed show
that negative content (eWOM/social media) is detrimental to sales (Chevalier and Mayzlin 2006, Sun 2012).
However, other studies have shown that even the presence of negative content increases product evaluations and
sales (Doh and Hwang 2009, Hiura et al. 2009). Generally the mere availability of peer opinions has a positive
influence on consumers, regardless of whether these opinions are positive or negative (Godes and Mayzlin 2009).
In this context even negative sentiment can help increase brand awareness and potentially brand value. Therefore,
on the one hand the volume of negative (positive) user actions might be negatively (positively) related to brand
value; but on the other hand even negative user actions can have a positive relationship with brand value. In any
case, according to Mazin (2011), over half of Facebook users exclusively share positive content about brands,
with the rest sharing a mixture of both positive and negative content, meaning that user sentiment is typically
for the most part positive on Facebook (Schweidel and Moe 2014). Thus, intuitively we argue that user actions
should be positively related to brand value. We thus formulate the following hypotheses:
Hypothesis 1 (H1). Brand actions on Facebook are positively related to brand value.
Hypothesis 2 (H2). User actions on Facebook are positively related to brand value.
Hypothesis 3 (H3). Brand actions on Facebook have are positively related to user actions on Facebook.
Brand and User Actions on YouTube. After brand and user actions on Facebook our next constructs
represent brands’ and users’ respective level of engagement on YouTube. This is reflected by subscriptions to
the brand’s YouTube channel as well as by the number of videos viewed. Users are typically more responsive
to images and videos than written text (Turri et al. 2013), potentially making YouTube a powerful marketing
medium. Brands typically use YouTube to create awareness, prescribe a solution, or spread information quickly
(Liu-Thompkins and Rogerson 2012). To do so brands typically post videos of product reviews, demonstrations,
unboxing of products, and brand-related events (Smith et al. 2012); all potentially positively associated with
brand value. However, as will be outlined in the research methodology section, systematically collecting data
about such actions proved impossible at the time of the study, and thus, this construct ultimately had to be
omitted.
Having set up a YouTube channel, brands must acquire subscribers and encourage them to watch their branded
videos (Liu-Thompkins and Rogerson 2012). Over half (52%) of consumers say that watching product videos
makes them more confident in online purchase decisions (Dusto 2012). We argue that the extensive audience
that YouTube offers, along with the media richness of the channel, can be used by brands to generate increased
brand value, leading to the following hypothesis:
Hypothesis 4 (H4). User actions on YouTube are positively related to brand value.
Brand and User Actions on Twitter. Twitter was primarily developed as a tool to promote conversation
(Kietzmann et al. 2011) and an estimated 19% of all tweets are brand related (Jansen et al. 2009). Brand actions
come in the form of tweets posted by the brand on its own Twitter account, which helps foster user awareness
(Toubia and Stephen 2013) and potentially leads to word-of-mouth (Jansen et al. 2009). Posting such content
is a direct way to attract followers and foster retweets from users. As with Facebook activity, we expect that

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
156 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Figure 1. Conceptual Model
"RANDACTIONSON
&ACEBOOK
5SERACTIONSON
9OU4UBE
"RANDACTIONSON
4WITTER
5SERACTIONSON
4WITTER
(
(
(
5SERACTIONSON
&ACEBOOK
(
(
(
(
"RANDVALUE
brand activity on Twitter will also drive user activity. Building up a pool of followers allows brands to establish
a direct connection with their user base, facilitating the flow of information and influence (Liu-Thompkins and
Rogerson 2012). This can help drive brand awareness and change attitudes among prospective customers, as
well as contribute to the acquisition of new customers (Malthouse et al. 2013), thus positively affecting brand
value. Followers can also retweet brands’ tweets, spreading the brands’ message further afield.
Along the same lines, an active base of followers can also potentially influence brand performance. For
example, Yadav et al. (2013) investigate the relationship between Twitter followers and sales, finding that active
Twitter usage can influence pre-purchasing decisions as well as future sales. Others, such as Culnan et al.
(2010), found that a high number of followers and retweets leads to higher firm performance. Finally, Kumar
and Mirchandani (2012) conducted an isolated study in which they found that 23% of the revenues from a
specific marketing campaign could be attributed to Twitter conversations. As a result, we believe that Twitter
has the potential to be positively related to brand value. Thus, see the hypotheses below:
Hypothesis 5 (H5). Brand actions on Twitter are positively related to brand value.
Hypothesis 6 (H6). Brand actions on Twitter are positively related to user actions on Twitter.
Hypothesis 7 (H7). User actions on Twitter are positively related to brand value.
Brand Value
In this study, we make use of both the Interbrand and Brand Finance approaches to brand valuation. Interbrand
has been widely used in previous research as a proxy for brand value (see, e.g., Barth et al. 1998, Madden
et al. 2006). Their approach uses a brand-earnings multiplier to calculate brand value, with weights calculated
from historical data such as brand share and advertising expenditures, as well as individuals’ judgments of other
factors such as brand stability and international reputation (Kapferer 2008). Brand value is the product of this
multiplier and the average of the past three years’ profits.
Notwithstanding Interbrand’s reputation, in our study we decided to supplement it with the Brand Finance
brand value metric to give a more comprehensive estimate of brand value. Brand Finance’s metric is regarded as
a valid alternative to Interbrand (Haigh and Gilbert 2005) and thus combining both approaches helps us obtain
a more complete estimate of brand value. Figure 1summarizes our conceptual model of how social media
contributes to brand value.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 157
Table 2. Brands in the Sample
Brand Region/Country Sector Brand Value 2013 ($m) Brand Value 2012 ($m) Aware (%)
3M United States Diversified 51413 41656 91
Accenture United States Business Services 91471 81745 87
Adidas Germany Sporting Goods 71535 61699 88
Adobe United States Technology 41899 41557 94
Allianz Germany Financial Services 61710 61184 92
Amazon United States Retail 231620 181625 95
AmericanExpress United States Financial Services 171646 151702 92
Apple United States Technology 981316 761568 92
Audi Germany Automotive 71767 71196 85
Avon United States FMCG 41610 51151 95
AXA France Financial Services 71096 61748 88
BMW Germany Automotive 311839 31922 92
Budweiser∗United States Alcohol 121614 291052 88
Burberry United Kingdom Luxury 51189 111872 90
Canon Japan Electronics 101989 41342 88
Cartier France Luxury 61897 121029 91
Caterpillar United States Diversified 71125 51495 97
Chevrolet∗∗∗ United States Automotive 41578 Not included 95
Cisco United States Technology 291053 61306 90
Citi United States Financial Services 71973 271197 95
Coca-Cola United States Beverages 791213 71570 93
Colgate United States FMCG 71833 771839 93
Corona∗Mexico Alcohol 41276 71643 96
Danone France FMCG 71968 31866 90
Dell United States Technology 61845 71498 96
Discovery∗∗∗ United States Media 51756 Not included 92
Disney United States Media 281147 71591 92
Duracell∗∗∗ United States FMCG 41645 Not included 95
eBay United States Retail 131162 271438 72
Facebook United States Technology 71732 101947 95
Ferrari Italy Automotive 41013 51421 92
Ford United States Automotive 91181 31770 89
Gap United States Apparel 31920 71958 80
GE United States Diversified 461947 31731 95
Gillette United States FMCG 251105 431682 95
GoldmanSachs United States Financial Services 81536 241898 95
Google United States Technology 931291 71599 93
Gucci Italy Luxury 101151 691726 49
H&M Sweden Apparel 181168 91446 93
Harley-Davidson United States Automotive 41230 161571 90
Heineken∗Netherlands Alcohol 41331 31857 88
Heinz United States FMCG 71648 31939 88
Hermes France Luxury 71616 71722 88
Honda Japan Automotive 181490 61182 93
HP United States Technology 251843 171280 93
HSBC United Kingdom Financial Services 121183 261087 96
Hyundai South Korea Automotive 91004 111378 93
IBM United States Business Services 781808 71473 88
IKEA Sweden Retail 131818 751532 80
Intel United States Technology 371257 121808 88
Design and Methodology
Sample Selection
The intended population for this study was Interbrand’s 100 most valuable brands. Other researchers in market-
ing have used this Interbrand brand value sample (see, e.g., Torres et al. 2012). However, certain brands had to

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
158 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Table 2. (Continued)
Brand Region/Country Sector Brand Value 2013 ($m) Brand Value 2012 ($m) Aware (%)
J.P.Morgan United States Financial Services 111456 391385 97
JackDaniel’s∗United States Alcohol 41642 111471 95
JohnDeere United States Diversified 41865 41352 95
JohnnieWalker∗United Kingdom Alcohol 41745 41221 79
Johnson&Johnson United States FMCG 41777 41301 89
Kellogg’s∗∗ United States FMCG 121987 41378 88
KFC United States Restaurants 61192 121068 53
Kia South Korea Automotive 41708 51994 95
Kleenex United States FMCG 41428 41089 94
L’oreal France FMCG 91874 41360 95
LouisVuitton France Luxury 241893 81821 90
MasterCard∗∗∗∗ United States Financial Services 41206 231577 92
McDonald’s United States Restaurants 411992 31896 92
Mercedes-Benz Germany Automotive 311904 401062 94
Microsoft United States Technology 591546 301097 96
MorganStanley United States Financial Services 51724 571853 96
Moet&Chandon∗France Alcohol 31943 71218 90
MTV United States Media 41980 31824 95
Nescafe Switzerland Beverages 101651 51648 90
Nestle Switzerland FMCG 71527 111089 73
Nike United States Sporting Goods 171085 61916 95
Nintendo Japan Electronics 61086 151126 85
Nissan Japan Automotive 61203 71082 95
Nokia Finland Technology 71444 41969 75
Oracle United States Technology 241088 211009 96
Pampers United States FMCG 131035 221126 96
Panasonic Japan Electronics 51821 111296 94
Pepsi United States Beverages 171892 51765 89
Philips Netherlands Electronics 91813 161594 89
PizzaHut United States Restaurants 41269 91066 96
Porsche Germany Automotive 61471 41193 85
Prada Italy Luxury 51570 51149 94
RalphLauren United States Apparel 41584 41271 95
Samsung South Korea Technology 391610 41038 86
Santander∗∗∗∗ Spain Financial Services 41660 321893 75
SAP Germany Technology 161676 41771 91
Shell Netherlands Energy 51535 151641 87
Siemens Germany Diversified 81503 41788 72
Smirnoff∗United Kingdom Alcohol 41262 71534 94
Sony Japan Electronics 81408 41050 92
Sprite United States Beverages 51811 91111 95
Starbucks United States Restaurants 41399 51709 92
ThomsonReuters Canada Media 81103 41062 92
Tiffany&Co. United States Luxury 51440 81444 10
Toyota Japan Automotive 351346 51159 95
UPS United States Transportation 131763 301280 76
Visa United States Financial Services 51465 131088 92
Volkswagen Germany Automotive 111120 41944 86
Xerox United States Business Services 61779 91252 79
Zara Spain Apparel 101821 31851 86
Notes. The % of “aware” represents the number of people that are aware of the brand. The data was provided by YouGov
brand.
∗excluded because of age restrictions on social media brand pages access.
∗∗excluded because of absence of an official Facebook presence.
∗∗∗excluded because of differences in the composition of the 2012 and 2013 Interbrand brand value lists.
∗∗∗∗excluded because of the absence of any social media data.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 159
be removed because of age restrictions on social media brand pages access (Budweiser, Corona Extra, Heineken,
Jack Daniel’s, Johnnie Walker, Moet and Chandon, Smirnoff), absence of an official Facebook presence (Kel-
logg’s), differences in the composition of the 2012 and 2013 Interbrand brand value lists (Chevrolet, Discovery,
and Duracell), and the absence of any social media data (MasterCard and Bank Santander). The final sample
size was thus composed of 87 brands. The brands included in the study are described in Table 2.
Data
Social Media Measures. As explained in the conceptual background section we separate social media actions
into two distinct categories—brand actions and user actions. To measure these actions we collected past con-
versations on three targeted social media platforms (Facebook, Twitter, and YouTube). Since the social media
space is vast and constantly changing (Peters et al. 2013), coverage of the entire spectrum of available systems
was not possible. Although company blogs and user review sites—both dedicated as well as within retail sites
such as Amazon.com or Expedia.com—have in the past been major avenues for social media activity, such sites
have already been widely investigated in academic research (see, e.g., Tirunillai and Tellis 2012, Luo et al.
2013b). In addition the social media space has evolved, with the majority of interactions having moved to the
three most popular generalist social media platforms, namely Facebook, Twitter, and YouTube. Facebook counts
over 1.4 billion users with 4.5 billion likes per day (Statista 2015) although nearly two out of three (58%)
Facebook users have liked a brand (Mazin 2011). Twitter counts 288 million active users, who send 500 million
tweets per day (Twitter 2014), with over one-third of Twitter users (36%) following at least one brand (Edison
Research 2014). Finally, YouTube counts over 1 billion users (YouTube 2015) with over 300 hours of videos
uploaded every minute (YouTube 2015). Such usage statistics highlight the importance of these three social
media platforms, making them relevant for investigation.
Since obtaining historical data directly is problematic, we sourced our data from a third-party provider that
collects and archives social media data using a set of automated tools. We ascertained the validity of these
data by accessing each brand’s social media asset (Facebook page, Twitter account, YouTube channel) over a
Figure 2. Interbrand Brand Value
3EGMENTSARETYPICALLYDEFINEDBYGEOGRAPHYBUSINESSUNITPRODUCTSERVICEORCUSTOMERGROUP
&INANCIAL!NALYSIS%CONOMIC0ROFIT$EMAND!NALYSIS2OLEOF"RAND)NDEX #OMPETITIVE!NALYSIS"RAND
3TRENGTH3CORE
"RAND%ARNINGS
"RANDV
"RAND2ISK
ALUE
"RANDVTRENGTHMEASURESTHE
ABILITYOFTHEBRANDTOCREATELOYALTY
ANDTHEREFORETOKEEPGENERATING
DEMANDANDPROFITINTOTHEFUTURE
&INALLYrole of brand index IS
multiplied by the ECONOMICPROFIT
of the branded PRODUCTSOR
services to determine THE
earnings attributable to the BRAND
(“brand earnings ) that CONTRIBUTE
3EGMENTATION
%CONOMICPROFITISTHEAFTERTAXOPERATING
PROFITOFTHEBRANDMINUSACHARGEFOR
THECAPITALUSEDTOGENERATETHEBRANDS
REVENUESANDMARGINS
2OLEOFBRANDindex MEASURESTHEPORTIONOF
the PURCHASEDECISIONTHATISATTRIBUTABLETO
THEBRANDRELATIVETOOTHERFACTORSEG
PURCHASEDRIVERSLIKEPRICECONVENIENCE
ORPRODUCTFEATURES
.ETPRESENTVALUEOFBRANDEARNINGS
!PROPRIETARYFORMULAISUSEDTO
CONNECTTHEbRANDsTRENGTHsCORE
TOABRANDSPECIFICDISCOUNTRATE
h
'LVFRXQW5DWH)
to the valuation total.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
160 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Table 3. Descriptive Statistics 4N =875
Obs. with
Constructs Metric missing data∗Mean Std. deviation
Brand actions Number of photos posted 2 379084 311013
on Facebook Number of videos posted 2 56008 111045
Number of status updates 2 42068 58057
Number of miscellaneous posted 2 18007 33071
User actions Number of likes/brands’ posts 1 218561604010 5395812026
on Facebook Number of comments/brands’ posts 1 761631070 1101904050
Number of shares/ brands’ posts 1 2311890020 4681520053
Number of PTAT 2 515541359078 811541479000
Number of likes on brands’ page 0 213831807001 319041886067
Brand actions Number of tweets 4 31691028 91212074
on Twitter
User actions Number of followers 4 5131880073 110491981065
on Twitter Number of retweets 1 561893028 1811722026
User actions Number of subscribers 4 501137033 1501665037
on YouTube Number of video views 3 1115101221081 2014671923018
Brand value Brand value Interbrand 0 1125313901804060 3185317501194028
Brand value Brand Finance 8 1125012531164056 3169313031904082
Notes. All variables are operationalized as a difference between December 2013 and December 2012 value.
∗Missing values were treated implicitly in the algorithm, as PLS-PM algorithm does not require any explicit missing
values substitution, using the Lohmöller (1989) mean method.
10-day period and manually collected each of our study metrics. We then compared our observations with the
data provided by the vendor and found no discrepancies, indicating that data was being reliably archived from
Facebook, Twitter, and YouTube.
Brand Value. Brand values are published by Interbrand and Brand Finance at the end of each year. We
recorded our social media metrics at the end of 2012 and 2013 to be able to relate activity to brand value in the
same years. We present a detailed explanation of Interbrand’s brand value approach in Figure 2.
We collected data twice to be able to operationalize variables as the difference between the 2012 and 2013
values. Therefore, each variable measures an increase or decrease in the absolute value of each metric and thus
describes the nature of the phenomenon in 2013. This was done in order to understand how a change in each
social media metric could relate to a change in brand value. Descriptive statistics are provided in Table 3. The
variables in our model have different dispersions and measurement units, and thus we standardized the data by
subtracting the variables’ mean and dividing by their standard deviation.
Model
Partial least squares path modeling (PLS-PM) is a multivariate technique that allows researchers to simultane-
ously estimate a series of interrelated dependence relationships (Tenenhaus et al. 2005). It has been previously
used in marketing (see, e.g., Henseler et al. 2009) as well as in other fields (see, e.g., Peng and Lai 2012) and
was selected for use in this study for a variety of reasons. First, in contrast to covariance based approaches
to structural equation (CBSEM) modeling, PLS-PM is a component based estimation algorithm that explains
the residual variance of both the construct and the indicators (Vinzi et al. 2010), an objective well-suited to
this study as we wished to assess how social media activities explain the variability in brand value. Second,
PLS-PM is more conservative that CBSEM as it underestimates path coefficients and avoids improper solutions
that sometimes occur in CBSEM (Tenenhaus 2008). Third, we performed both univariate and multivariate tests
for normality and established that the assumption of multivariate normality is violated, thus further justifying
the use of PLS-PM for its soft distributional assumptions (Chin 2010). Finally, we have a limited sample size
(87) and research agrees that in order to achieve the same consistency of parameter estimation CBSEM requires
a sample size of several orders of magnitude larger than PLS-PM (Chin 2010).
Results
Based on the guidelines provided in Chin (2010) the evaluation of the measurement model and the structural
model are presented below.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 161
Table 4. Inter-Construct Squared Correlations and Reliability Measures the Constructs
Composite Brand actions User actions Brand actions User actions User actions Brand
reliability AVE Construct on Facebook on Facebook on Twitter on Twitter on YouTube value
0.87 0.62 Brand actions
on Facebook
1.00
0.90 0.63 User actions
on Facebook
0.16 1.00
∗ ∗ Brand actions
on Twitter
0.30 0.01 1.00
0.85 0.69 User actions
on Twitter
0.15 0.35 0.04 1.00
0.88 0.72 User actions
on YouTube
0.00 0.01 0.00 0.01 1.00
0.87 0.76 Brand value 0.01 0.01 0.00 0.00 0.43 1.00
∗single item construct.
Measurement Model. The model was examined to insure that constructs are well related to their indicators.
To assess reflective indicators we examined internal consistency, convergent validity, and discriminant validity
(Chin 2010). Internal consistency was assessed by examining the composite reliability index (Fornell and Larcker
1981), which Hair et al. (2013) suggest that it should be higher than 0.7, a threshold exceeded by all constructs
Table 5. Loadings and Cross-Loadings for the Measurement (Outer) Model
Constructs/ Brand actions User actions Brand actions User actions User actions
Indicators on Facebook on Facebook on Twitter on Twitter on YouTube Brand value
Number of photos
posted
0083 0042 0049 0027 −0008 −0006
Number of status
updates
0082 0034 0056 0055 −0003 0013
Number of videos
posted
0073 0024 0027 0008 0011 0016
Number of
miscellaneous
posted
0076 0018 0032 0026 −0007 0019
Number of likes on
brands’ page
0030 0069 0012 0031 0008 −0001
Number of PTAT 0037 0091 0008 0045 0008 −0003
Number of
likes/brands’ posts
0023 0079 0007 0042 0001 −0007
Number of
comments/brands’
posts
0039 0083 0037 0077 0010 −0011
Number of
shares/brands’
posts
0016 0072 −0004 0021 0003 −0004
Number of tweets 0054 0011 1000 0021 −0005 −0002
Number of followers 0020 0059 0007 0065 0022 0005
Number of retweets 0040 0052 0022 0098 0007 −0005
Number of
subscribers
−0007 0007 −0006 0013 0098 0071
Number of video
views
0009 0009 0000 0004 0069 0026
Brand value
Interbrand
−0002 −0009 −0004 −0002 0072 0094
Brand value Brand
Finance
0030 −0001 0003 −0003 0036 0080
Note. The items in bold represent the loadings of the items on their respective constructs.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
162 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Table 6. Results of the Structural Model
Path Path coefficient Tstat. Hypothesis Result
Brand actions on Facebook →Brand value 0027 2073∗∗∗ H1Supported
User actions on Facebook →Brand value −0021 −2039∗∗ H2Not supported
Brand actions on Facebook →User actions on Facebook 0040 3097∗∗∗ H3Supported
User actions on YouTube →Brand value 0069 8066∗∗∗ H4Supported
Brand actions on Twitter→Brand value −0028 −1.12ns H5Not supported
Brand actions on Twitter→User actions on Twitter 0021 1098∗H6Supported
User actions on Twitter→Brand value −0009 −0.80ns H7Not supported
Notes. Values shown are standardized coefficients. The items in bold represent the significant T-statistics.
ns =nonsignificant; ∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001 level.
in our model by a large margin (see Table 4). Similarly, each indicator’s loading should be higher than 0.707
(squared loading of 0.5) meaning that at least half of the item variance is extracted by its respective construct
(see Table 5). Most loadings exceed this proposed threshold, with a small number borderline (0.69 for the
number of likes on brand page, 0.65 for number of followers, 0.69 for number of videos views). Hair et al. (2013)
recommend that loadings between 0.4 and 0.7 should be further evaluated for their significance and content
validity. All borderline loadings were significant at 95% level and are important in terms of content validity
established in accordance to our theoretical framework. Thus, we have retained all of the original items in our
model.
Convergent validity was evaluated by assessing the outer loadings and using the average variance extracted
(AVE) criterion as suggested by Fornell and Larcker (1981). An acceptable value is greater than or equal to 0.5
(Fornell and Larcker 1981). This is exceeded for each of our constructs as shown in Table 4, implying good
convergent validity.
Finally, discriminant validity indicates the extent to which a construct is different from other constructs (Chin
2010). There are two ways to assess discriminant validity. First, the AVE should be higher than the squared
correlations among the constructs. In our model, all constructs have a higher correlation with their measurement
items than with any other construct (see Table 4). Discriminant validity can also be assessed by examining
the cross-loadings, where the indicator loading should be lower than their loadings on any other construct (see
Table 5). In our model, all indicators load higher on their own constructs than on any other construct, indicating
good discriminant validity. Overall, our model shows good internal consistency, convergent and discriminant
validity, and we can therefore proceed to evaluating the structural model.
Structural Model. In the structural model the most important criterion for goodness of prediction is the
amount of variance explained in the endogenous constructs, which in our model is equal to 0.495. We therefore
assert that changes in social media metrics (reflected by our selected theoretical constructs) explain 49.5% of
the variance of the change in brand value, which we find highly satisfactory.
In Table 6we present the structural model results. All path coefficients are standardized and should be
interpreted in terms of standard deviation. The significance of these relationships is assessed by bootstrapping
analysis (Tenenhaus et al. 2005). Chin (2010) suggests that path values should not be lower than 0.20, a criterion
respected in the model results. Hypotheses H1, H3, H4, and H6are thus supported, with H5(p-value =00425)
and H7(p-value =00224) not supported. Furthermore, we find the opposite effect to what was expected for H2.
We find strong support for our H1and H4. The exogenous constructs brand actions on Facebook (=0027,
p < 0001) and user actions on YouTube (=0069, p < 00001) exert a positive influence on brand value. We
also find support for our Hypothesis H3of the positive association between brand actions on Facebook on user
actions on Facebook (=0040, p < 0001) and moderate support of our Hypothesis H6of the positive association
between brand actions on Twitter and user actions on Twitter (=0021, p < 001).
However, we do not find support for the association between either user actions on Twitter or brand actions
on Twitter, respectively, and brand value (H5and H7). Instead, we find both constructs are negatively (although
not significantly) related to brand value. Similarly with H2we, surprisingly, find a negative and significant
(= −0021, p < 0005) coefficient, implying that user actions on Facebook has a negative effect on brand value.
Further Analysis
To provide further insights we conducted complementary PLS-PM analysis (mediation, moderation, and hetero-
geneity) in subsequent steps after the original analysis of the PLS path model (as discussed in Hair et al. 2013).

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 163
Table 7. Direct, Indirect, and Total Effects in the Structural Model with the Inclusion of WOM
Brand value
Path WOM Direct Indirect
Brand actions on Facebook −0007ns 0035∗∗∗ −0009ns
User actions on Facebook 0028∗∗ −0029∗∗ 00123∗∗
User actions on YouTube 0027∗∗ 0059∗∗∗ 0011∗∗
Brand actions on Twitter −0015ns −0017ns −0008ns
User actions on Twitter 0008ns −0006ns 0003ns
WOM — 0042∗∗∗ —
Note. Values shown are standardized coefficients.
ns =nonsignificant; ∗∗p < 0005; ∗∗∗ p < 0001 level.
Mediation. Our initial model solely considered the direct effects of social media on brand value. In this
section we propose a mediation analysis which provides further insights into the relationship between social
media actions and brand value. Having analysed the literature, we identified a list of potential mediators and
focused our attention on the most important one, word-of-mouth (WOM), performing additional analysis to
establish its role in mediating the relationship between social media and brand value.
WOM is an important factor that affects consumers’ purchase and adoption decisions (Trusov et al. 2009), as
well as brand sales (Chevalier and Mayzlin 2006, You et al. 2015). In terms of word-of-mouth’s antecedents,
Berger and Schwartz (2011) have found that products which are more publicly visible typically receive more
WOM, both immediately and over time. As engaging in social media activity increases visibility, we expect that
social media can drive word-of-mouth that in turn drives the value of the brand.
We obtained data on word-of-mouth spread from YouGov Group, which uses online consumer panels to
measure brand perceptions. YouGov monitors multiple brands by periodically surveying a panel of over 5 million
consumers. To assure representativeness YouGov weights the sample by age, race, gender, education, income,
and geography. In any single survey respondents are only asked about one measure for each sector to ensure
that none of the survey metrics influence each other, thus reducing common method bias and measurement error.
The YouGov database has been previously used in marketing literature (e.g., Luo et al. 2013a) and presents a set
of advantages. First, YouGov administers the same set of questions for each brand, assuring consistency across
brands for each of their metrics. Second, the YouGov database is considered reliable as it uses a large panel
of consumers, capturing the general opinion of the crowd. Lastly, the large panel size and random selection of
respondents implies that YouGov data coherently captures between-subject variance (Luo et al. 2013b).
YouGov provided us with WOM metrics for both December 2012 and 2013, specifically as to whether panel
members had talked about each brand with friends and family in the previous two weeks (whether in-person,
online, or through social media). To test for mediation effects of WOM we assess whether social media is related
to WOM and if WOM is related to brand value. We first assessed the measurement model (discriminant and
convergent validity) and, as in the initial analysis, obtain satisfactory results. In the structural model the variance
in the change in brand value explained becomes 60.1%, implying that WOM explains an additional 10.6% of
variance with respect to the previous model. We present the results of the structural model with the inclusion of
WOM in Table 7. We find that user actions on Facebook and user actions on YouTube are significantly related
to WOM, and that WOM significantly relates to brand value, providing preliminary evidence for the mediating
role of WOM.
To formally assess the mediating role of word-of-mouth, we follow common practice in marketing and other
fields and use the Sobel (1982) test. In line with current developments in mediation analysis (see Zhao et al.
2010), we do not make distributional assumptions on the effects and use the bootstrapped standard errors (from
5,000 bootstrapped resamples) to compute the t-value for the Sobel test, finding evidence that WOM mediates
the relationship between user actions on YouTube and brand value (Sobel test statistic =20029, p-value <0005).
With the inclusion of WOM as a mediator, the direct effect of user actions on YouTube on brand value remains
positive and significant. Both effects (direct and indirect) point to the same direction, providing evidence for
the partial and complementary mediation of WOM (see, e.g., Zhao et al. 2010). We also find evidence that
WOM mediates the relationship between user actions on Facebook and brand value (Sobel test statistic =10962,
p-value <0005). Again, with the inclusion of WOM as mediator the direct relationship between user actions on
Facebook and brand value remains negative and significant. In this case, however, the indirect effect points in
the opposite direction, making the case for competitive and partial mediation. Thus, we find that user actions

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
164 Service Science 8(2), pp. 152–168, © 2016 INFORMS
on Facebook have a dual effect: the direct effect on brand value is negative (we postulate because of negative
sentiment contained in these user actions) whereas the indirect effect is positive due to WOM spread. Once we
account for this competitive mediation, the effect of user actions on Facebook on brand value remains negative
but are reduced to −0.168. Finally, we do not find mediation evidence for the relationship between brand actions
on Facebook and brand value (Sobel test = −00492, p-value =00622).
Alternative Mediator: Awareness. To further investigate the possible alternative explanations for the effect
of social media on brand value we also investigated another potential mediator—awareness. Social media can be
used by brands to create awareness (Liu-Thompkins and Rogerson 2012) which enables consumers to include
brands in their consideration set. It is well established that awareness is an important first step in the brand
purchase funnel (Murdough 2009) and potentially can affect the value of the brand. Therefore, we assess whether
brand awareness could mediate the relationship between social media and brand value.
We tested this hypothesis by requesting additional data on brand awareness from YouGov. YouGov provided
us with data of whether panel members were familiar with the brand. We test whether awareness mediated the
significant direct relationships between our social media variables and brand value. As before, we first assessed
the measurement model (discriminant and convergent validity) and obtained satisfactory results. However, we do
not find any significant relationship between social media and awareness, nor between awareness and brand value
(results available upon request). We postulate that this may be because most brands in the top 100 Interbrand
list are already well-known brands, and thus, most consumers are already aware of their existence (see Table 2).
As a result, it is not surprising that awareness does not mediate the relationship between social media and brand
value in our study.
Moderation. Another alternative explanation of the effects in our model could be related to the differences
among brands within the sample. For example, stronger brands could have larger social media presences that
attract more user actions and thus have a stronger relationship with brand value. To test moderation in our model,
we used multiple group analysis (Eberl 2010). We obtain data from YouGov on brand strength and perform a
median split on this variable to separate the sample into comparatively stronger and less strong brands (above
and below the median brand strength score, respectively). We obtained a discrete moderator variable that can be
interpreted as dividing the data into two subsamples. Next, we estimated the same PLS-PM for each of the two
distinct subsamples. We used Dibbern and Chin’s (2005) distribution-free approach, using a random permutation
procedure to assess the differences among groups. We set the number of permutations to 500 and compared
the path coefficients between two groups at a time, which allows an interpretation of the differences in effects
between groups. We did not find any significant differences in terms of path coefficients in the model (results
available upon request), and conclude that brand strength does not moderate the relationship between social
media and brand value. We did, however, suspect that some alternative moderators (other than brand strength)
could be relevant for our model. To partially address this issue we checked for the presence of heterogeneity in
our sample.
Heterogeneity. PLS-PM assumes homogeneity over the observed set of units, implying that all units are
represented by a single model estimated on the overall data set. Accordingly, we checked for the presence of
both observed and unobserved heterogeneity in the sample. This analysis was carried out to rule out alternative
explanations for the results in our model. If the sample turned out to be heterogeneous it might be possible
that some underlying brand characteristics influenced the results and we could not generalize our findings to all
brands in the sample. On the other hand, if we find homogeneity across brands, we can be relatively sure that
the results from our model are robust and apply to all the brands in our sample.
Observed heterogeneity. We checked whether the mean of the standardized scores from the PLS-PM varied
for segments of brands based on latent variables’ scores. This analysis is useful to detect observed heterogeneity
which could indicate the existence of a moderating variable. In particular we perform a median split on all the
latent variables in the model to separate the sample into two comparable groups (above and below the median
latent variable score). Next we estimated the same PLS path model in each of the two distinct groups, using
analysis of variance to investigate the differences among these groups. Our results suggest that the mean of the
latent variables does not indicate any observed heterogeneity in the model (results are available upon request).
Unobserved heterogeneity. Once we tested for a possible moderator (brand strength) and for observed het-
erogeneity, the last step was to assess the unobserved heterogeneity, which implies identifying classes of units
(a priori unknown) having similar behaviors. Such heterogeneity is captured by an unobserved discrete moder-
ating variable defining both the number of classes and the class membership (Vinzi et al. 2008). We used the
response-based procedure for detecting unit segments in PLS-PM (REBUS-PLS) (Vinzi et al. 2008) that does
not require distributional hypotheses. We also used the alternative approach for unobserved heterogeneity, the

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 165
finite mixture partial least squares (FIMIX-PLS) (Hair et al. 2013)—a segmentation that captures heterogeneity
by estimating the probabilities of segment memberships for each observation and simultaneously estimating the
path coefficients of all segments. Both algorithms indicated the absence of unobserved heterogeneity allowing
us to safely conclude that our sample is homogenous.
Thus, both our moderation and heterogeneity analyses indicate that brands in our sample are quite similar
and we can safely conclude that our results are robust to the inclusion of control variables and hold for all the
brands analysed.
Discussion and Implications
As a new and developing field, social media is potentially well suited for brand marketing actions, with marketers
well aware of the potential. For most brands, social media presents an interesting opportunity for both keeping
in contact with customers as well as growing brand value. However, the key question is how to justify the
investments needed to meaningfully engage in social media marketing (Deans 2011). On a more granular level
brands also need to understand which of the many possible social media actions influence brand value, positively
or negatively, so as to be able to identify which should be prioritised and more clearly focus their efforts.
To address these questions we examined a broad range of brand and user social media actions on Twitter,
Facebook, and YouTube, building a theoretical framework to assess how these metrics interact with brand value.
We subsequently validated our model using data from the social media profiles of the top 100 brands as ranked
by Interbrand. This empirically demonstrated that changes in certain social media actions are associated with a
change in brand value for the 2013 year, thus demonstrating the effect of social media on brand value.
Our proposed theoretical model was found to have a high level of explanatory power, explaining 49.5% of
the variance in brand value.User actions on YouTube and brand actions on Facebook were found to have the
strongest positive influence on brand value, with an increase of one standard deviation in each construct resulting
in an increase of 0.69 and 0.29 standard deviations, respectively, in brand value. In contrast, user actions on
Facebook were found to have a negative influence on brand value. An increase of one standard deviation in
this construct resulted in a decrease of 0.21 standard deviations in brand value. A possible explanation for this
can be found in Smith et al. (2012), who state that social media can potentially convey and amplify negative
sentiment as it promotes conversation and link sharing. Overall, evidence on the impact of negative content is
inconclusive (Babi´c et al. 2015). To investigate this further, we tested for the mediating role of word-of-mouth.
We found that by including WOM in the model, its overall explanatory power raises to 60.1%. We do find that
user actions on Facebook are significantly related to a mediator, word-of-mouth, translating in a positive and
significant indirect effect on brand value. Therefore, despite the progress in this paper, we encourage future
researchers to investigate under what condition user actions on Facebook can be related to brand value either
positive and negatively. Additionally, user actions on YouTube have a significant effect both on WOM and brand
value, implying that WOM partially mediates this relationship.
To assess the robustness of the model we conducted moderation and heterogeneity analysis. We found that the
results of our model are not sensitive to the inclusion of brand strength as a moderator. Although the brands in
this study are mostly strong, analysing how their relative strength affects the results provides additional evidence
for the robustness of our model. Our analyses did not find any differences in terms of model parameters for the
strong and less strong brands’ groups. We also did not find any observed or latent heterogeneity in the brands
studied. Thus, both our moderation and heterogeneity analyses suggest that brands in our study are quite similar
and we can conclude that our results are robust to the inclusion of control variables and hold true for the brands
analysed, making a case for the role of social media actions in the brand value of the Interbrand top 100 brands.
Overall, our findings indicate that brands should consider using YouTube as a content community (see, e.g.,
Smith et al. 2012), focusing on driving both the number of subscribers and the number of video views. In
addition brands should be active on Facebook, posting photos, video, and status updates frequently to gain
attention from users, although whether they should encourage user reaction is questionable since our findings
suggest that such actions are negatively related to brand value. However, our mediation analysis indicates that
user actions can still have a positive effect on brand value through word-of-mouth spread and future research
should look into the conditions under which (e.g., valence, industry setting) such user actions could be related
to brand value.
Lastly, our model also examined whether brand actions influence user actions on the same platform. We
identified strong support for a positive association between brand actions on Facebook and user actions on
Facebook, as well as moderate support for a positive association between brand actions on Twitter and user
actions on Twitter. This implies that brands have the ability to stimulate and shape user conversation on social
media.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
166 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Our findings give a first insight into how to quantify and assess the effect of social media actions on branding.
To our knowledge this is the first study that attempts to quantitatively address this important and topical issue.
As such it helps advance the marketing literature in terms of understanding the effect of making use of this
developing medium on branding, and thus, makes important theoretical and practical contributions. From a
theoretical perspective the study proposes and validates a comprehensive set of objective metrics to help measure
social media actions, a first step in theory building on social media marketing. It also empirically tests the
association of changes in these metrics with brand value, something currently lacking in the marketing literature.
Lastly, it proposes and tests a framework to measure this relationship using real-world data from social media
networks, discovering that brand actions on YouTube and Facebook have the strongest association with brand
value.
From a managerial perspective our model makes an important contribution as we have built and validated
a model that answers the following question:Is investment in social media really worth it? We demonstrate
that certain social media actions are typically associated with increased brand value, a first step in justifying
investments in social media marketing. In addition, our findings provide insight into how marketers should
approach the use of social media platforms to maximise their positive effect on brand value.
As with most pieces of research, our study suffers from several limitations. First, we examined only the top
100 Interbrand brands and do not attempt to generalize beyond this population. Such brands obviously do not
represent the entire spectrum of brands available, and thus, future studies could include more brands or analyse
particular industry sectors singularly in order to provide deeper insights. Second, notwithstanding the fact that
Facebook, Twitter, and YouTube are currently the largest social media platforms, further research might take
into consideration a broader range or an alternative set of platforms to help validate and increase the reliability
of these results. Lastly, our study suffers from omitted variable bias as it was not possible to collect historical
data on brand actions on YouTube, and thus, what we recognise as an important variable could not be included
in our model.
Acknowledgments
The authors thank YouGov for providing BrandIndex data to facilitate the study.
References
Babi´c A, Sotgiu F, de Valck K, Bijmolt THA (2015) The effect of electronic word of mouth on sales: A meta-analytic review of platform,
product, and metric factors. J. Marketing Res. Forthcoming.
Barth ME, Clement MB, Foster G, Kaznik R (1998) Brand values and capital market valuation. Rev. Acounting Stud. 3(1):41–68.
Berger J, Schwartz EM (2011) What drives immediate and ongoing word of mouth? J. Marketing Res. 48(5):869–880.
Bruhn M, Schoenmueller V, Schäfer DB (2012) Are social media replacing traditional media in terms of brand equity creation? Management
Res. Rev. 35(9):770–790.
Chen Y, Liu Y, Zhang J (2012) When do third-party product reviews affect firm value and what can firms do? The case of media critics
and professional movie reviews. J. Marketing 76(2):116–134.
Chevalier JA, Mayzlin D (2006) The effect of word of mouth on sales: Online book reviews. J. Marketing Res. 43(3):345–354.
Chin W (2010) How to write up and report PLS analyses. Vinzi V, Chin W, Henseler J, Wang H, eds. Handbook of Partial Least Squares.
Springer Handbooks of Computational Statistics (Springer, Berlin), 655–690.
Chintagunta PK, Gopinath S, Venkataraman S (2010) The effects of online user reviews on movie box office performance: Accounting for
sequential rollout and aggregation across local markets. Marketing Sci. 29(5):944–957.
Culnan MJ, Mary J, McHugh PJ, Zubillaga JI (2010) How large U.S. companies can use Twitter and other social media to gain business
value. MIS Quart. Executive 9(4):243–259.
Deans PC (2011) The impact of social media on C-level roles. MIS Quart. Executive 10(4):187–200.
Dellarocas C (2006) Strategic manipulation of internet opinion forums: Implications for consumers and firms. Management Sci. 52(10):
1577–1593.
Dibbern J, Chin WW (2005) Multi-Group Comparison: Testing a PLS Model on the Sourcing of Application Software Services Across
Germany and the USA Using a Permutation Based Algorithm. Handbuch PLS-Pfadmodellierung. Methode, Anwendung, Praxisbeispiele,
135–160.
Doh S-J, Hwang J-S (2009) How consumers evaluate eWOM (electronic word-of- mouth) messages. Cyberpsychology Behav. 12(2):193–197.
Duan W, Gu B, Whinston A (2008) The dynamics of online word-of-mouth and product sales—An empirical investigation of the movie
industry. J. Retailing 84(2):233–242.
Dusto A, (2012) Online videos help build confidence in purchases. Invodo (April 5), https://www.internetretailer.com/2012/04/05/online
-videos-help-build-confidence-purchases.
Eberl M (2010) An application of PLS in multi-group analysis: The need for differentiated corporate-level marketing in the mobile communi-
cations industry. Vinzi V, Chin W, Henseler J, Wang H, eds. Handbook of Partial Least Squares, Springer Handbooks of Computational
Statistics (Springer, Berlin), 487–514.
Edison Research (2014) The social habit 2014. Accessed March 3, 2015, http://www.edisonresearch.com/social-habit-report/.
Fornell C, Larcker DF (1981) Evaluating structural equation models with unobservable variables and measurement error. J. Marketing Res.
18(1):39–50.
Godes D, Mayzlin D (2009) Firm-created word-of-mouth communication: Evidence from a field test. Marketing Sci. 28(4):721–739.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
Service Science 8(2), pp. 152–168, © 2016 INFORMS 167
Goh KY, Heng CS, Lin Z (2013) Social media brand community and consumer behavior: Quantifying the relative impact of user-and
marketer-generated content. Inform. Systems Res. 24(1):88–107.
Haigh D, Gilbert S (2005) Valuing not-for-profit and charity brands—Real insight or just smoke and mirrors. Internat. J. Nonprofit Voluntary
Sector Marketing 10(2):107–119.
Hair JF, Hult GTM, Ringle CM, Sarstedt M (2013) A Primer on Partial Least squares Structural Equation Modeling 4PLS-SEM5(Sage
Publications, Thousand Oaks, CA).
Heggestuen J, Danova T (2013) Brand presence: How to choose where to be on social media. Business Insider (August 18),
http://www.businessinsider.com/social-media-strategy-for-brands-2013-8?IR=T.
Henseler J, Ringle CM, Sinkovics RR (2009) The use if partial least squares path modeling in international marketing. Adv. Inter nat.
Marketing 8(20):277–319.
Hiura K, Kikumori M, Kishimoto K, Matsumoto N, Nakagawa M, Ujita M, Ono A (2009) Does negative e-WOM affect consumer attitude
negatively or positively? J. eBusiness 10(1):38–42.
Ho-Dac NN, Carson SJ, Moore WL (2013) The effects of positive and negative online customer reviews: Do brand strength and category
maturity matter? J. Marketing 77(6):37–53.
Hoffman DL, Fodor M (2010) Can you measure the ROI of your social media marketing? MIT Sloan Management Rev. 52(1):41–49.
Jansen BJ, Zhang M, Sobel K, Chowdury A (2009) Twitter power: Tweets as electronic word of mouth. J. Amer. Soc. Inform. Sci. Technol.
60(11):2169–2188.
Kapferer JN (2008) Strategic Brand Management, New Approaches to Creating and Evaluating Brand Equaty (Free Press, New York).
Keller KL, Lehmann DR (2006) Brands and branding: Research findings and future priorities. Marketing Sci. 25(6):740–759.
Kietzmann JH, Hermkens K, McCarthy IP, Silvestre BS (2011) Social media? Get serious! Understanding the functional building blocks of
social media. Bus. Horizons 54(3):241–251.
Kim AJ, Ko E (2012) Do social media marketing activities enhance customer equity? An empirical study of luxury fashion brand. J. Bus.
Res. 65(10):1480–1486.
Kumar V (2015) Evolution of marketing as a discipline: What has happened and what to look out for. J. Marketing 79(1):1–9.
Kumar V, Mirchandani R (2012) Increasing the ROI of social media marketing. MIT Sloan Management Rev. 54(1):55–61.
Liu-Thompkins Y, Rogerson M (2012) Rising to stardom: An empirical investigation of the diffusion of user-generated content. J. Interactive
Marketing 26(2):71–82.
Lohmöller JB (1989) Latent Variable Path Modeling with Partial Least Squares (Physica-Verlag, Heidelberg, Germany).
Luo X, Raithel S, Wiles MA (2013a) The impact of brand rating dispersion on firm value. J. Marketing Res. 50(3):399–415.
Luo X, Raithel S, Wiles MA, Zhang J, Duan W (2013b) Social media and firm equity value. Inform. Systems Res. 24(1):146–163.
Madden TJ, Fehle F, Fournier S (2006) Brands matter: An empirical demonstration of the creation of shareholder value through branding.
J. Acad. Marketing Sci. 34(2):224–235.
Malthouse EC, Haenlein M, Skiera B, Wege E, Zhang M (2013) Managing customer relationships in the social media era: Introducing the
social CRM house. J. Interactive Marketing 27(4):270–280.
Mangold WG, Faulds DJ (2009) Social media: The new hybrid element of the promotion mix. Bus. Horizons 52(4):357–365.
Marketing Science Institute (2014) 2014–2016 research priorities. Accessed March 3, 2015, http://www.msi.org/research/2014-2016
-research-priorities/.
Mazin L (2011) Branding and how it works in the social media age. ATYM Blogs Accessed April 21, 2015, http://aytm.com/blog/research
-junction/branding-and-how-it-works-in-the-social-media-age/.
McCann U (2013) Cracking the code: The story of why. Universal McCann Wave 7. Accessed March 3, 2015, http://wave.umww.com/
assets/pdf/wave_7-cracking-the-social-code.pdf.
Mela CF, Gupta S, Jedidi K (1998) Assessing long-term promotional influences on market structure. Internat. J. Res. Marketing 15(2):89–107.
Motameni R, Shahrokhi M (1998) Brand equity valuation: A global perspective. J. Product Brand Management 7(4):275–290.
Murdough C (2009) Social media measurement: It’s not impossible. J. Interactive Advertising 10(1):94–99.
Nielsen (2012) State of the media: The social media report 2012. Accessed March 3, 2015, http://www.nielsen.com/us/en/insights/reports/
2012/state-of-the-media-the-social-media-report-2012.html.
Onishi H, Manchanda P (2012) Marketing activity, blogging and sales. Internat. J. Res. Marketing 29(3):221–234.
Peng DX, Lai F (2012) Using partial least squares in operations management research: A practical guideline and summary of past research.
J. Oper. Management 30(6):467–480.
Performics (2012) Life on demand study. Accessed March 3, 2015, http://blog.performics.com/life-on-demand-study-reveals-new-social
-norms-40-percent-of-americans-feel-more-comfortable-engaging/.
Peters K, Chen Y, Kaplan AM, Ognibeni B, Pauwels KH (2013) Social media metrics—A framework and guidelines for managing social
media. J. Interactive Marketing 27(4):281–298.
Schau HJ, Muñiz Jr AM, Arnould EJ (2009) How brand communities create value. J. Marketing 73(5):30–51.
Schweidel DA, Moe WW (2014) Listening in on social media: A joint model of sentiment and venue format choice. J. Marketing Res.
51(4):387–402.
Simon CJ, Sullivan MW (1993) The measurement and determinants of brand equity: A financial approach. Marketing Sci. 12(1):28–52.
Smith AN, Fischer E, Yongjian C (2012) How does brand-related user-generated content differ across YouTube, Facebook, and Twitter?
J. Interactive Marketing 26(2):102–113.
Sobel ME (1982) Asymptotic confidence intervals for indirect effects in structural equation models. Sociol. Methodology 13(1):290–312.
Srinivasan S, Hanssens DM (2009) Marketing and firm value: Metrics, methods, findings, and future directions. J. Marketing Res. 46(3):
293–312.
Statista (2015) Statistics and facts about social networks. Accessed March 3, 2015, http://www.statista.com/topics/1164/social-networks/.
Stephen AT, Galak J (2012) The effects of traditional and social earned media on sales: A study of a microlending marketplace. J. Marketing
Res. 49(5):624–639.
Sun M (2012) How does the variance of product ratings matter? Management Sci. 58(4):696–707.
Tenenhaus M (2008) Component-based structural equation modelling. Total Quality Management 19(7–8):871–876.
Tenenhaus M, Vinzi VE, Chatelin Y-M, Lauro C (2005) PLS path modeling. Computational Statist. Data Anal. 48(1):159–205.
Tirunillai S, Tellis GJ (2012) Does chatter really matter? Dynamics of user-generated content and stock performance. Marketing Sci.
31(2):198–215.

Colicev, O’Connor, and Vinzi: Is Investing in Social Media Really Worth It?
168 Service Science 8(2), pp. 152–168, © 2016 INFORMS
Torres A, Bijmolt THA, Tribó JA, Verhoef P (2012) Generating global brand equity through corporate social responsibility to key stake-
holders. Internat. J. Res. Marketing 29(1):13–24.
Toubia O, Stephen AT (2013) Intrinsic vs. image-related utility in social media: Why do people contribute content to Twitter? Marketing
Sci. 32(3):368–392.
Trusov M, Bucklin RE, Pauwels KH (2009) Effects of word-of-mouth versus traditional marketing: Findings from an internet social
networking site. J. Marketing 73(5):90–102.
Turri AM, Smith KH, Kemp E (2013) Developing affective brand commintment through social media. J. Electronic Commerce Res.
14(3):201–214.
Twitter (2014) By the numbers: 116 amazing Twitter statistics. Accessed March 3, 2015, https://twitter.com.
Vinzi VE, Trinchera L, Amato S (2010) PLS path modeling: From foundations to recent developments and open issues for model assessment
and improvement. Vinzi V, Chin W, Henseler J, Wang H, eds. Handbook of Partial Least Squares. Springer Handbooks of Computational
Statistics (Springer, Berlin), 47–83.
Vinzi VE, Trinchera L, Squillacciotti L, Tenenhaus M (2008) REBUS PLS: A response based procedure for detecting unit segments in PLS
path modelling. Appl. Stochastic Models Bus. Indust. 24(5):439–458.
Weinberg BD, Pehlivan E (2011) Social spending: Managing the social media mix. Bus. Horizons 54(3):275–282.
Yadav MS, de Valck K, Hennig-Thurau T, Hoffman DL, Spann M (2013) Social commerce: A contingency framework for assessing
marketing potential. J. Interactive Marketing 27(4):311–323.
You Ya, Vadakkepatt GG, Joshi AM (2015) A meta-analysis of electronic word-of-mouth elasticity. J. Marketing 79(2):19–39.
YouTube (2015) YouTube statistics. Accessed March 3, 2015, http://www.youtube.com/yt/press/statistics.html.
Yu Y, Duan W, Cao Q (2013) The impact of social and conventional media on firm equity value: A sentiment analysis approach. Decision
Support Systems 55(4):919–926.
Zhao X, Lynch JR, Chen Q (2010) Reconsidering Baron and Kenny: Myths and truths about mediation analysis. J. Consumer Res.
37(2):197–206.