Television Advertising and Online Word-of-Mouth: An …...1 Television Advertising and Online Word-of-Mouth: An Empirical Investigation of Social TV Activity Beth L. Fossen1 David
Post on 15-Oct-2020
0 Views
Preview:
Transcript
1
Television Advertising and Online Word-of-Mouth: An Empirical Investigation of
Social TV Activity
Beth L. Fossen1
David A. Schweidel2
April 2015
1 Doctoral Candidate in Marketing, Goizueta Business School, Emory University. Address: 1300
Clifton Road Northeast, Atlanta, Georgia 30322, Email: bfosssen@emory.edu
2 Associate Professor of Marketing, Goizueta Business School, Emory University. Address: 1300
Clifton Road Northeast, Atlanta, Georgia 30322, Email: dschweidel@emory.edu
The authors thank Kantar Media and the Goizueta Business School research support for providing the
funds to acquire and build the dataset in this research.
2
Television Advertising and Online Word-of-Mouth: An Empirical Investigation of
Social TV Activity
Abstract
In this research, we investigate how television advertising drives online word-of-mouth
(WOM). We first explore if television advertising (1) affects online WOM about the brand
advertised and (2) associates with changes in online WOM about the program in which the
advertisement airs. We further examine if the media context in which the advertisement appears
– the television program – impacts the relationship between television advertising and online
WOM. By investigating the integration of consumer social media participation with television
programming, known as social TV, we aim to improve the field’s understanding of the consumer
experience with television, advertising, and social media.
Using data containing television advertising instances and the volume of minute-by-minute
social media mentions, our analyses reveal that television advertising impacts online WOM for
both the brand advertised and the program in which the advertisement airs. We additionally find
that the programs that receive the most online WOM aren’t necessarily the best programs for
advertisers in terms of online engagement. These results suggest the need for social TV activity
to be viewed in terms of viewer engagement with both programs and advertisements. Moreover,
the results indicate that the program in which the advertisement airs affects the extent of online
WOM for both the brand and program following television advertising. Overall, this research
sheds light on how marketers, television networks, and program creators can (1) increase online
WOM for their respective brands and programs through media planning and advertisement
design strategies and (2) incorporate online WOM into the media planning and buying process.
Keywords: online word-of-mouth, advertising, television, social media, social TV
3
1. Introduction
Does television advertising drive online word-of-mouth (WOM)? Is an advertisement’s ability to
generate online WOM determined by the brand advertised, or does the context in which the
advertisement airs – the television program – also play a role? Furthermore, does television
advertising have the ability to affect the volume of online conversations about the brand
advertised and the program in which the advertisement airs? Research in the marketing literature
has established that online WOM matters and can increase new customer acquisition (e.g.,
Trusov et al. 2009), television ratings (e.g., Godes and Mayzlin 2004), and sales (e.g., Chevalier
and Mayzlin 2006; Moe and Trusov 2011; Kumar et al. 2013; Rishika et al. 2013). While the
positive consequences of online WOM have been identified, research into the drivers of online
WOM is still in its infancy, and the extant literature has yet to explore these questions. As the
rise of multi-screen media consumption has outpaced the field’s understanding of the activity
(e.g., Hare 2012; Poggi 2012; Copeland 2013), the relationship between television advertising
and online WOM is an increasingly important topic to investigate in order to assess the extent to
which marketers can leverage such multi-screen behavior.
In this research, we address these questions and investigate how television advertising
contributes to online conversations, bridging the gap in extant literature on online WOM and
television viewing. Specifically, we explore how primetime television advertising contributes to
online WOM about (1) the brand featured in the advertisement and (2) the television program in
which the advertisement airs. We further examine how specific brand, advertisement, and
program characteristics impact online brand and program chatter and assess the potential value
of social TV activity, defined as social media interaction with television programming (Hill et al.
2012), to marketers.
4
Nielsen (2014) estimates 84% of tablet and smartphone users engage in multi-screen
behavior while watching television. Joint work from Twitter, FOX, and the Advertising Research
Foundation finds that 85% of Twitter users active during primetime programming contribute to
online conversations about television and 90% of users exposed to this chatter have taken TV-
related action such as switching channels to watch a program or searching online for additional
program information (Midha 2014). Overall, the global media industry’s interest in social TV is
substantial as social-media related television businesses comprised a $151 billion industry in
2012; this number is estimated to grow to $256 billion in 2017 (Lomas 2012). Despite this rapid
growth in social TV activity, advertisers and networks are facing challenges trying to grasp the
value of this behavior (e.g., Hare 2012; Poggi 2012; Copeland 2013). With this study, we
contribute to research on online WOM and television advertising by advancing understanding of
social TV behavior and exploring the value of this activity for both advertisers and networks.
We construct a data set that includes television advertising instances on network
broadcasts, minute-by-minute social media data of Twitter conversations mentioning brands and
programs, and data on brand, advertisement, and program characteristics. We supplement this
data by coding the use of calls-to-action in each advertisement (e.g., does an ad feature a
hashtag?) to assess their impact on online WOM. Our data include over 9,000 ad instances for
264 brands across 15 categories that aired on 84 primetime network programs during the fall
2013 television season1.
We jointly model the immediate change in online mentions for both the brand and the
program following an advertisement’s airing. To assess if the context in which a television ad
airs influences its impact on online WOM, we incorporate a measure of brand-program fit based
on advertisers’ choices of which programs to air their ads during (Schweidel et al. 2014), which
1 Additional details of our data are presented in Tables A1 and A2 in the Online Appendix.
5
we refer to as brand-program synergy. We use this model framework to evaluate the impact of
brand and advertisement characteristics (e.g., calls-to-action) and program characteristics (e.g.,
genre, network) on online WOM. In doing so, our analysis sheds light on how advertisers and
television networks can encourage online WOM for their respective brands and programs.
We find evidence of increases in online mentions for both brands and programs following
advertisements, illustrating that television advertising has the potential to increase online WOM
about the advertised brand and the program in which the ad airs. This result reveals the potential
benefit of social TV for advertisers and also challenges the industry perspective that television
viewers are less likely to engage in online program chatter during commercials (Nielsen 2013).
We also find that the increase in online WOM from television advertising depends on brand-
program synergy, with online WOM increasing more for both brands and programs when there is
high synergy. This suggests that the context in which an advertisement airs influences online
WOM and could have implications for advertiser-network negotiations as it provides an
incentive for both parties to consider brand-program synergies in the media buying process.
Interestingly, we also find that brands that advertise in programs that experience higher
than expected online program chatter following television advertisements don’t necessarily
experience increases in online WOM for the brand. This suggests that the programs that receive
the most online WOM aren’t necessarily the best programs for advertisers seeking to generate
online chatter. Our results also shed light on the factors that influence social TV activity as we
discover that numerous brand, advertisement, and program characteristics impact online brand
and program WOM following television advertising. Of note, we find that advertisements
featuring digital calls-to-action can increase immediate online WOM for the advertised brand.
This effect, however, only occurs if the advertisement is shown in the first ad slot in a
6
commercial break. These results have implications for advertisement design strategies and the
importance that ad position should play in media buy negotiations for advertisers interested in
online WOM.
The remainder of this research is organized as follows. In the next section, we briefly
review related literature on online WOM as well as research on television consumption and
advertising. We then describe the data, present model-free evidence, and discuss the modeling
approach for jointly assessing the immediate impact of television advertising on online WOM for
brands and programs. We present our results and conclude with a discussion of implications of
our research, as well as opportunities for future research, in the contexts of online WOM, social
TV, and the media planning process.
2. Background Literature
2.1 Online WOM
Why would marketers, television networks, and program creators want to encourage online
WOM? Recent research on online WOM illustrates that online conversations matter. For
example, consumer participation in social media has been shown to increase sales and create a
positive return on investment (Kumar et al. 2013) and increase shopping frequency and
profitability (Rishika et al. 2013). Additional studies have shown that other forms of online
WOM, such as online reviews and ratings (Chevalier and Mayzlin 2006; Moe and Trusov 2011)
and social earned media from online communities and blogs (Stephen and Galak 2012), can also
impact sales. Online WOM has additionally been linked to increased online customer acquisition
(Trusov et al. 2009). A recent Nielsen survey with participants from 56 countries supports these
findings, discovering that 46% of survey respondents used social media to inform purchase
7
decisions (Nielsen 2012). Further research on online WOM also illustrates the value of online
chatter to television networks and program creators. Godes and Mayzlin (2004) show that online
WOM activity relates to television program ratings. A recent investigation by Gong et al. (2014)
also finds that Tweets from a program’s microblog account and influential re-Tweets can
increase program ratings. Industry reports from Kantar Media and Nielsen complement this
research by illustrating that Twitter activity during television broadcasts correlates with higher
program ratings (Subramanyam 2011; Kantar Media 2014).
Online WOM also can serve as a proxy for brand engagement. Hoffman and Fodor
(2010) argue that marketers can use customer social media interactions to measure engagement.
Recent research has operationalized various forms of online WOM as measures of engagement,
such as comments and replies on social networking websites like Facebook and Twitter (Kumar
et al. 2013; Rishika et al. 2013). This suggests that online conversations can serve as an indicator
that an advertisement (or program content) is engaging.
While the above literature highlights the importance of online WOM for marketers,
research on how the consumer generation of online WOM can be encouraged is still in its early
stages. Work in this area by Berger and Schwartz (2011) finds that more WOM is generated for
products that are publicly visible or cued by the surrounding environment. Research has also
begun to explore the drivers of online WOM including how content affects individuals’ decisions
to share news articles online (Berger and Milkman 2012). Additional investigations have
examined the dynamics in consumer decisions to contribute to online content (e.g., Godes and
Silva 2012; Moe and Schweidel 2012). This research, however, has not considered social TV
behavior. Thus, we have limited insight into how marketers, television networks, and program
creators can increase online WOM for their respective brands and programs, and our
8
understanding of the consumer experience with television, advertising, and social media remains
incomplete.
2.2 Television Consumption and Advertising
In addition to work on online WOM, our research also relates to investigations of television
consumption and advertising. Television viewing is still the most prominent form of media
consumption in the U.S. with Americans watching on average five hours of television a day
(Katz 2013). Marketers spent over $64 billion on television advertising in the U.S. in 2012 with a
30 second advertisement on primetime network broadcasts ranging in costs from $24,000 to
$545,000 with an average cost of around $121,000 (eMarketer 2013; Katz 2013).
Why might television advertising impact online WOM both for the brand advertised and
for the program in which the advertisement airs? Recent research on cross-media effects has
presented initial evidence that television advertising can influence online behavior as it has been
found to impact branded keyword search (Joo et al. 2014), shopping behavior in terms of website
traffic and online sales (Liaukonyte et al. 2015), and blogging activity for movies and wireless
phone services (Onishi and Manchanda 2012). Additionally, Gopinath et al. (2014) explore the
relationship between total advertising instances across all channels, online WOM, and brand
performance. They find initial evidence that advertising in one month can impact online WOM
in the next month. This research, however, does not consider the integration of consumer social
media participation with television programming.
Online WOM also may be influenced by television advertising in a more direct manner if
an advertisement includes a call-to-action. For example, an advertisement featuring a hashtag, a
keyword used to contribute to an online conversation about a topic, may increase online brand
WOM by informing viewers that an online dialog exists. Alternatively, featuring an offline call-
9
to-action such as a toll-free number may have the reverse effect, decreasing online brand WOM
by encouraging an offline action. This view is consistent with research on direct response
advertising that has shown that toll-free numbers in television advertisements can significantly
increase the number of incoming calls, an offline response (e.g., Tellis et al. 2000). In addition to
online WOM for brands, calls-to-action in television advertisements also may impact online
WOM for programs, as consumers that respond to an advertisement’s call-to-action may be less
likely to engage in online discussion about the program. We empirically examine these
possibilities and assess what impact various calls-to-action have on online WOM for both brands
and programs.
Television advertising also may influence online WOM for both brands and programs
because commercial breaks are natural pauses in program content, and viewers may be more
likely to engage in online conversations during natural breaks (Dumenco 2013). However,
Nielsen (2013) argues that television viewers are actually less likely to engage in online program
chatter during commercials. We empirically explore these contrasting viewpoints and assess if
television advertising is associated with increases in online WOM for both brands and programs.
Why might the program in which an advertisement airs influence the ad’s impact on
online WOM? Literature on media context effects evaluating the effect of media engagement on
advertising response has generally found that more engagement with a television program relates
to improved consumer response to the advertising (e.g., Murry et al. 1992; Feltham and Arnold
1994; Tavassoli et al. 1995; Wang and Calder 2009). While this literature does not address
online WOM and generally focuses on program engagement rather than advertisement
engagement, it suggests that the media context should matter when assessing consumer response
10
to advertising. Recent investigations in the marketing literature have called for further research
on how television content and television advertising interact (Wang and Calder 2009).
We expand upon this work on television consumption and advertising by exploring social
media interactions with television viewing to gain a broader understanding of the consumer
experience with television, advertising, and social media. Additionally, we contribute to research
on media context effects by extending this body of work into the online WOM context, exploring
online engagement with advertisements, and evaluating the potential interaction between
advertisement and television program WOM in social TV behavior.
3. Data Description
3.1 Television Advertising Data
Data on primetime television advertising during the fall 2013 television season (Sept. 1 – Dec.
31, 2013) were gathered from Kantar Media’s Stradegy database. Data were collected for
advertisements that aired in primetime (8pm-11pm) on broadcast networks (ABC, CBS, CW,
FOX, and NBC) during the initial airing of recurring programs2. We exclude advertisements that
are joint promotions, ads that feature two or more brands, from our analysis3. Our final data set
of television advertisements consists of 9,103 ad instances for 264 brands across 15 categories
that aired on 84 television programs.
3.2 Social Media Data
We combine the television advertising data with minute-by-minute level data of brand and
program mentions on Twitter. The data were accessed using Topsy Pro which, at the time that
data were collected, was a certified Twitter partner with access to the Twitter Firehose of
2 Recurring programming does not include programs that only air once, such as the Miss America competition. 3 Joint promotions made up less than 1% of advertising instances in our data.
11
comprehensive, real-time Twitter posts dating back to 20064. We focus our analysis of social TV
activity on Twitter posts because the majority of public social media chatter about television
occurs on Twitter5. We utilize the number of online mentions for brands and programs to assess
online WOM. A search of program mentions on Twitter was conducted by capturing Tweets that
contain a program’s name or nickname (e.g., Parks and Recreation, “Parks and Rec”), hashtags
featuring a program’s name or nickname (e.g., #parksandrecreation, #parksandrec,
#parksandrecnbc), or the program’s Twitter handle (e.g., @parksandrecNBC)6. We employed a
similar strategy to search brand mentions on Twitter, capturing Tweets that mention the brand, a
hashtag featuring the brand name, a hashtag included in the brand’s advertisement, or the brand’s
Twitter handle. Parent brand names were used as the focus of the search over product brand
names (e.g., “Colgate” versus “Colgate Optic White”) for almost all brands. The exceptions
include motion pictures (e.g., parent brand – Warner Brothers; product brand – Gravity), books
(e.g., parent brand –Little, Brown And Company; product brand – Gone by James Patterson),
tech products (e.g., parent brand – Amazon; product brand – Kindle), and brands that share a
name with a common word (e.g., Coach, Halls, Nationwide). For these exceptions, product brand
names were incorporated into the search of Twitter brand mentions. Tables A4 and A5 in the
Online Appendix present search terms used to capture brand mentions for a subsample of the 264
brands in our analysis.
3.3 Data on Brand, Advertisement, and Program Characteristics
We supplement the data set of television advertising instances and online WOM with additional
brand characteristics. We control for the category of the advertised brand as certain categories
4 Topsy Pro was acquired by Apple in late 2013 and is no longer available to the public. 5 Bluefin Labs estimates 95% of public social media chatter about television occurs on Twitter (Schreiner 2013). 6 Note that this search strategy does not double count conversations that include more than one of these elements.
12
may vary in their ability to generate online WOM7. We also account for if a brand does not have
a Twitter profile as it may reflect a brand’s indifference toward online WOM and may influence
consumer decisions to discuss the brand online. We additionally include a number of
advertisement characteristics in our analysis. Most notably, we code the use of calls-to-action in
each advertisement, indicating if an ad contains a phone number, Facebook page link or icon,
hashtag, and/or web address. We also control for ad length and ad position both in a commercial
break (first, middle, or last non-promo ad) and in a program (ad break position and if an ad airs
near a half-hour interval). This data were extracted from Stradegy. Lastly, we account for if an
advertisement runs in a simultaneous commercial break with another ad instance by the same
brand. We define simultaneous as an advertisement for the same brand airing within two minutes
(either before or after) of the advertisement of interest. This time window was chosen in
correspondence with our dependent variable, which will be discussed in the following section.
Finally, we account for program characteristics as they may influence how an
advertisement impacts online WOM. We control for network, program genre, Nielsen program
rating, day of the week the program airs, and time the program airs. This data were gathered
from the Stradegy database with the exception of program ratings which were collected from TV
by the Numbers. We further control for season premiere and fall finale episodes as these episodes
may generate more social chatter compared to other episodes.
3.4 Descriptive Statistics
Tables A1 and A2 in the Online Appendix display advertising instances by category, all 84
programs explored in this analysis, and the most advertised brands by category and program
genre. The most advertised categories are movies, beauty, and wireless providers which account
7 Table A1 in the Online Appendix shows the 15 categories utilized in our analysis as well as the number of
advertisement instances and top advertisers in each category.
13
for nearly 40% of the ad instances in the data. The most advertised brands are Apple, Microsoft,
AT&T, Nokia, Sprint, and Bank of America which account for 20% of the ad instances in our
sample. Table 1 shows summary statistics for the brand, advertisement and program
characteristics. Of note, the majority of the advertising instances in our sample (65%) are 30
seconds. Only 1% of ad instances are longer than 30 seconds in length. Additionally, 59% of the
advertisement instances in our data feature a web address, 17% include a hashtag, 6% contain a
phone number, and 5% include a Facebook page link or icon. We also see that nearly half of the
ad instances in our data (47%) air on Drama/Adventure programs, as about half of the programs
in our data are classified in this genre (45%). Lastly, the average Nielsen program rating for
episodes in our sample is 1.85, and the majority of ads (62%) air on episodes with ratings
between 1.00-2.49. Figure A1 in the Online Appendix shows the frequency of advertisement
instances by program rating, day of the week, and time block.
[Insert Table 1 about Here]
3.5 Model-free Evidence
Can Television Advertising Impact Online WOM about the Brand? Our model-free evidence
suggests that television advertising can increase online WOM for the advertised brand,
illustrating that social TV activity may benefit marketers. Figure 1 shows the average brand
mentions per minute on Twitter for brands in several categories from thirty minutes prior to the
advertisements’ airings to thirty minutes after the ads air8. This figure consistently shows
increases in online brand mentions immediately following the advertisements’ airings. This
increase in per minute mentions spikes very quickly, around two minutes after the
advertisements air. The lifts in online WOM from television advertising vary across categories
with movies on average increasing their Twitter mentions by more than ten mentions per minute
8 In all the model-free evidence figures, the bold, dashed center line indicates when the advertisement airs.
14
following an advertisement while other categories see smaller lifts. Across all brands explored in
our analysis, the largest average increases in online WOM for brands two minutes after an
advertisement airs (compared to two minutes prior) occur for movie brands with Mandela: Long
Walk to Freedom, Insidious 2, and Lee Daniels’ The Butler seeing increases of 329, 175, and 122
mentions per minute, respectively.
[Insert Figure 1 about Here]
We also use model-free evidence to explore how online WOM for brands may be
impacted by key advertisement and program characteristics. Figure 2 provides evidence that
using calls-to-action in advertisements may influence online WOM for brands and that the
magnitude of these effects may be driven by ad position in a commercial break. Utilizing
hashtags and web addresses in an advertisement that airs first or in the middle of a commercial
break appears to positively impact online brand WOM compared to using these calls-to-action in
an ad that airs in the last slot. Figure 2 also illustrates how program ratings may influence an
advertisement’s effect on online brand chatter, suggesting that ads airing in programs with higher
ratings generally receive a greater lift in online WOM compared to ads airing in programs with
lower ratings. This is most notable by the large post advertisement spikes in per minute Twitter
mentions for brands airing in programs with ratings above 2.00.
[Insert Figures 2 about Here]
Can Television Advertising Impact Online WOM about Programs? We find model-free
evidence that television advertising can encourage increases in online WOM for programs.
Figure 3 shows the average program mentions per minute on Twitter around the airings of
advertisements and illustrates that, across all advertisements in our data, per minute program
chatter spikes shortly after the ads air. The largest spikes are seen following advertisements that
15
air in the first position of a commercial break. In our data, the largest average increases in online
program WOM two minutes after an ad airs (compared to two minutes prior), regardless of ad
position in the commercial break, occur for ABC’s Scandal, FOX’s Glee, and CW’s Vampire
Diaries, which see average increases of 276, 120, and 111 in per minute mentions, respectively.
These findings are consistent with the viewpoint that television viewers are more likely to
engage in online WOM during commercial breaks, possibly because they are natural pauses in
programming content (Dumenco 2013), and challenges the belief advocated by Nielsen (2013)
that television viewers are not reserving online program chatter for commercial breaks.
[Insert Figures 3 about Here]
Does Context Matter? Figure 4 provides model-free evidence that the media context in
which a television advertisement airs influences the ad’s impact on online WOM and helps
motivate the need to account for potential brand-program synergy in our formal model. For
example, Maybelline on average sees increases in their online brand chatter following the airing
of their television advertisements on ABC’s Scandal and FOX’s GLEE but on average
experiences decreases in online brand WOM following their ads airing on ABC’s 20/20. As
another example, Microsoft sees immediate increases in online WOM after their advertisements
air on FOX’s GLEE but see immediate decreases in online brand mentions following their ads
airing on ABC’s 20/20 and ABC’s Scandal. Figure 5 also presents evidence that the relationship
between brand and program matters when assessing the effect of television advertising on online
WOM. This figure shows the impact of advertisements by Sprint and the movie Gravity on
online brand WOM across different programs that aired on Sept. 23, 2013, and illustrates that the
impact varies based on the program in which the brand chooses to air the advertisement.
[Insert Figures 4 and 5 about Here]
16
Overall, as suggested by Figures 1-5, we see evidence of substantial heterogeneity in both
online brand and program WOM across various brand, advertisement, and program
characteristics. However, as this model-free evidence does not account for multiple factors and
does not broadly explore the synergistic association between online brand and program WOM
following the airing of advertisements, a formal model is needed to explore the relationship
between television advertising and online WOM. Toward this end, we next describe our
modeling framework.
4. Model Development
4.1 Joint Model of Online WOM about Brands and Programs
We jointly model the immediate change in online brand and program mentions following the
airing of a particular television advertisement. We measure this immediate change using narrow
two-minute windows before and after the advertisement airs. We specify the dependent variables
of interest for our primary analysis as follows:
(1)
,|eBWOM|PostBWOM
,PreBWOMPostBWOMY
ii
iii
)1Prlog(
)1log(1
0
0
ii
ii
PreBWOMPostBWOM
PreBWOMPostBWOM
where i indexes the ad instance in our data, PostBWOMi is the number of online Twitter
mentions for the brand in ad instance i two minutes after the advertisement airs, and PreBWOMi
is the number of online mentions for the brand two minutes before the advertisement airs. We
specify Yi2 in the same manner with PostPWOMi and PrePWOMi, the number of online Twitter
mentions for the program corresponding to ad instance i two minutes after and two minutes
before, respectively, the advertisement airs. We use log specifications as there is large variance
in these differences across brands. We attribute the differences between these pre- and post-
measures to the ad insertion. Narrow event windows have been employed in past research to
17
investigate the effects of television advertising on online behavior (e.g., Hill et al. 2012;
Liaukonyte et al. 2015). Furthermore, our model-free evidence consistently illustrates that almost
all of the immediate change in online WOM for both brands and programs occurs within two
minutes of the advertisement’s airing. Thus, we utilize the two-minute windows both before and
after the advertisement airs to assess the impact of television advertising on online WOM.
We consider a number of alternative specifications of the change in online mentions
brought about by television advertising. Specifically, we evaluate alternative models that
consider dependent variables that are not constructed as difference measures, account for WOM
valence, and utilize different pre- and post-advertisement time windows. The substantive results
for these alternative specifications do not differ from those of our main model. Detailed
discussions for these supplementary analyses are provided in the Online Appendix.
This narrow event window is also advantageous as it is key to the identification strategy
and alleviates endogeneity concerns that advertisers or television networks could choose a
certain time window to air an advertisement in order to impact online WOM. In advertiser-
network media buy negotiations, the specificity of timing when an ad will air is limited to the
quarter-hour level, and this timing is not stipulated in the advertiser-network contracts, 80% of
which are completed in the May up-front market several months prior to the start of the fall
television season (Liaukonyte et al. 2015). Additionally, television networks commonly order
advertisements across commercial breaks at random (Wilbur et al. 2013). Therefore, it is not
plausible to time an advertisement to air during a specific four minute period to influence social
media activity.
We model the immediate change in brand b’s online mentions (Yi1) and program p’s
online mentions (Yi2) from television advertisement instance i as follows:
18
(2)
,
ˆ
ˆ~
2
1
2
1
i
i
i
i
Y
YN
Y
Y
(3)
58
1 2
1][],[
2
1
2],[
1],[
2],[
1],[
2
1
ˆ
ˆ
k
ikk
kipib
ip
ip
ib
ib
prog
brand
i
i XBPSynergyY
Y
where μbrand and μprog are respective intercepts. We account for brand-specific effects (αb.) and
program-specific effects (γp.) in each equation for the 264 brands and 84 programs in our data
and use them to explore (1) if a specific brand or program experiences changes in online WOM
following an advertisement’s airing and (2) potential cross-effects – that is if a specific brand
(program) influence online program (brand) WOM following an ad’s airing. Furthermore, the
brand- and program-specific effects αb. and γp. control for many of the potential unobservables
related to advertisers, networks, or program content that may influence Yi1 and Yi2 (e.g., audience
heterogeneity). Xi.is a vector of ad instance-specific brand, advertisement, and program
characteristics described in the data section above. Additionally, Xi. includes measures to explore
if ad break position in a program has a non-linear relationship with online WOM and if the
effects of calls-to-action on online chatter are influenced by ad position in a commercial break,
as this could impact the perceived time a viewer has to respond to a call-to-action (Danaher and
Green 1997)9. Finally, to account for potential correlation between the two dependent variables
of interest, we allow for contemporaneous covariance in Τ.
BPSynergybp is a latent measure of brand-program synergy that we construct to assess if
the media context (i.e., the television program) in which a brand’s advertisement airs interacts to
impact online WOM. BPSynergybp can be thought of as a brand-program interaction as it
assesses if the synergy or fit between brand b and program p influences online chatter for both b
and p above and beyond the main effects associated with the brand (αb.) and the program (γp.).
9 A summary of the variables in Xi. is presented in Table A3 in the Online Appendix
19
We construct BPSynergybp using a proximity model (Bradlow and Schmittlein 2000; Schweidel
et al. 2014) of the brands’ choices of in which programs to advertise. By jointly modeling the
advertisers’ program choices with equations (1)-(3), our modeling framework not only allows for
the construction of a measure of brand-program synergy, but it also recognizes that advertisers’
program selections may be non-random (Manchanda et al. 2004; Schweidel et al. 2014).
We assume that the probability that brand b advertises in program p is a function of the
latent distance between b and p in a Euclidean space (LatentDistbp). We let Zbp=1 if brand b
advertises in program p during the 2013 fall television season and 0 otherwise, and let the
probability that b advertises in p be specified as follows:
(4) b
bp
bpLatentDist
ZP
)(1
1)1(
If λb > 0, it follows that brand b is less likely to advertise in program p as LatentDistbp increases,
which we find to be the case. LatentDistbp is constructed as the Euclidean distance between brand
b’s location and program p’s location in a latent space and is specified as follows:
(5) 2
222
11 )()( pbpbbp PBPBLatentDist
where Bb1 (Pp1) and Bb2 (Pp2) specify the location of brand b (program p) in the two-dimensional
Euclidean space10. We use the estimates of these locations to construct brand-program synergy
measures BPSynergybp, which we assume to be inversely related to LatentDistbp:
(6) bp
bpLatentDist
BPSynergy1
4.2 Estimation
10 To avoid shifts and rotations of the axes during model estimation, we assume the locations for three brands in the
Euclidean space prior to estimation. We position brand 1 at the origin (B11=B12=0) to avert an axis shift, brand 2 on
the positive x-axis (B21>0, B22=0) to prevent rotation over the y-axis, and brand 3 such that B32>0 to avoid rotation
over the x-axis. The remaining brands and all programs locations are estimated from the data.
20
The above equations are estimated jointly using a Bayesian hierarchical regression and Markov
chain Monte Carlo techniques in WinBUGS (http://www.mrc-bsu.cam.ac.uk/bugs/). We assume
that ),0(~ 11 Nb , ),0(~ 22 Nb , ),0(~ 11 Np , and ),0(~ 22 Np , with diffuse inverse-
gamma priors for the variances. We specify μbrand, μprog, η., and θk. with diffuse normal priors,
and Τ with a diffuse inverse Wishart prior. Additionally, we assume that ),(~ Nb ,
),(~ 111 Bb BNB , ),(~ 222 Bb BNB , ),(~ 111 Pp PNP , and ),(~ 222 Pp PNP , with diffuse normal
priors for .B and .P and diffuse inverse-gamma priors for the variances. The above equations
are estimated from three independent chain runs of 40,000 iterations with the first 20,000
iterations discarded as a burn-in. Our inferences are based on the remaining 20,000 draws from
each chain. Model convergence is assessed through the time series plots of the posterior draws
for each parameter, and these plots provide evidence consistent with model convergence.
5. Results
5.1 Model Comparison
To assess the importance of accounting for both brand- and program-level effects as well as
brand-program synergy in our model of television advertising’s impact on online WOM, we
compare our proposed model to several alternative models in Table 2. Deviance information
criterion (DIC), a likelihood-based measure that penalizes more complex model specifications,
and the mean absolute error (MAE) are used to compare our proposed model to the alternative
specifications. Lower DIC and MAE indicate better model fit.
We first consider a baseline model that includes only the brand, advertisement, and
program characteristic variables in Xi. (Model 1). We then build upon this baseline model to
assess how model fit is impacted when only brand effects (Model 2) or only program effects
21
(Model 3) are included. Additionally, we consider a model in which no cross effects (Model 4)
are incorporated and models in which only brand cross effects (Model 5) or only program cross
effects (Model 6) are included. Finally, we consider a model in which BPSynergybp is withheld
from equation (3) (Model 7). Adding BPSynergybp to Model 7 gives us our proposed model
(Model 8).
The DIC and MAE estimates in Table 2 establish that accounting for brand- and
program-specific effects in our model of television advertising’s impact on online WOM
improves model fit. Furthermore, we find that a meaningful relationship does exist between
brand and program online WOM following an advertisement’s airing, as incorporating brand-
program synergy into Model 8 improves overall fit. Additionally, excluding cross effects – that is
not accounting for brand-specific (program-specific) effects in the model of advertising’s impact
on online program (brand) WOM – hurts model fit. These observations provide evidence that the
media context in which advertisements air matters when investigating the relationship between
television advertising and online WOM. As Model 8 is our best fitting model, we focus our
discussion on the results from this model estimation.
[Insert Table 2 about Here]
5.2 What Impacts Online WOM about Brands?
We begin by discussing the results pertaining to how brand and advertisement characteristics
affect online WOM about brands11. As seen in Table 3, we find that online brand chatter varies
across categories with movies experiencing larger increases in online WOM. Our results also
indicate that ad length affects subsequent online brand chatter, with shorter ads seeing less online
WOM. One potential explanation for why longer advertisements result in increases to online
11 Posterior mean estimates of the variance and heterogeneity parameters in equations (2)-(3) are presented in Table
A6 in the Online Appendix.
22
WOM about brands may be because they stand out to consumers; in our data, only 1% of ad
instances were longer than 30 seconds. Longer advertisements also may offer more opportunities
for consumers to be exposed to the brand name, which could also explain the increase in online
WOM that they generate. These potential explanations are consistent with Teixeira et al. (2012)
who find that ad length increases attention concentration and decreases zapping behavior.
[Insert Table 3 about Here]
We see that calls-to-action can impact an advertisement’s effect on online brand WOM.
Table 3 shows that including a phone number in a television advertisement actually reduces
subsequent online brand WOM. This call-to-action may decrease immediate online brand WOM
by encouraging an offline action (e.g., Tellis et al. 2000). In contrast, including a hashtag or web
address in an advertisement can successfully increase online WOM about the brand. But, this
effect only occurs when the advertisement airs in the first ad slot of a commercial break. This
suggests that digital calls-to-action can stimulate online brand engagement. The interaction with
ad position in a commercial break may occur because consumers feel as if they have more time
to respond to a digital call-to-action and engage in online WOM at the beginning of the
commercial break without interrupting program viewing (Danaher and Green 1997). This finding
highlights the importance of ad position in a commercial break for advertisers interested in
increasing online WOM about their brands.
Table 4 presents the results of the influence of program characteristics on online brand
WOM following television advertising. We find a positive relationship between program ratings
and an advertisement’s impact on online brand chatter. As higher program ratings indicate a
larger viewing audience that can be exposed to the advertised brand, this larger base may
23
contribute to more online WOM for the brand. We also see that advertisements airing on the CW
network, known for its younger audience base, see more subsequent online brand WOM.
[Insert Table 4 about Here]
5.3 What Impacts Online WOM about Programs?
The results in Table 3 show that brand and advertisement characteristics can impact
online program WOM following television advertising. We see evidence that online WOM about
programs varies across the categories of the brands that advertise in the program, with programs
experiencing higher (lower) levels of subsequent online program chatter following movie ads
(cable providers and non-profit/PSA ads). This result suggests that the selection of advertisers by
television networks has an impact on online program WOM. We also find that ad position in a
commercial break is associated with changes in online program mentions. More (less) online
program WOM is seen following advertisements that air in the first (last) ad slot of a commercial
break. Online program chatter following the first ad occurs early in the commercial break,
whereas such chatter following the last ad of a break occurs during the program content. Thus,
this finding is consistent with the argument that viewers are more likely to engage in online
WOM during commercials, possibly because they are natural pauses in programming (Dumenco
2013). Given the desire to avoid interrupting program viewing, such actions are more likely to be
taken earlier in a commercial break (Danaher and Green 1997). This finding further challenges
the argument made by Nielsen (2013) that television viewers are actually less likely to engage in
online program chatter during commercials as our results indicate that there is a substantial
increase in online program WOM two minutes following the first advertisement’s airing
(compared to two minutes prior).
24
We also find a non-linear relationship between ad break position in a program and online
program chatter following advertisements. We find that subsequent online WOM increases, but
at a decreasing rate, during advertisements that air in later ad breaks in the program. This finding
may reflect that online WOM increases as the program progresses but eventually tails off as
viewers become engrossed by the program content and disengage from their second screen.
Table 4 illustrates that program characteristics influence online program WOM following
television advertising. We see variation in online program WOM following advertisements
across program genres. Compared to Slice-of-Life/Reality programs, we see that Comedy,
Drama/Adventure, News/Magazine, and Suspense/Mystery/Police programs all experience more
online program chatter. These differences may reflect variations in characteristics of the program
in each genre or differences in characteristics of the viewers. In addition, we find variations
across broadcast networks with programs on FOX and NBC experiencing fewer online mentions
following advertisements (relative to programs on ABC). Finally, ads airing in season premieres
and fall finales are associated with more online program WOM than ads that air on episodes in
the middle of the season.
5.4 Brand-Program Synergy
From the model of advertisers’ program choices12, we find, as expected, that
> 0 based on the
95% higher posterior density (HPD) interval. This indicates that as the proximity between brand
b and program p increases, the probability that b advertises in p increases. As shown in Table 4,
we find that higher brand-program synergy is associated with increases in online WOM
following television advertising for both brands and programs. This suggests increased online
WOM for both the brand and program when they exhibit higher synergy. This illustrates that the
media context in which a television advertisement airs influences the ad’s impact on online 12 The full results from this model are shown in Table A7 in the Online Appendix.
25
WOM. Some examples of brand-program pairs with high synergy include Amazon-ABC’s
Grey’s Anatomy, Frozen (the movie)-FOX’s X Factor, Citibank-CBS’s Undercover Boss, and
Bank of America-CBS’s Amazing Race. Some examples of brand-program pairs with low
synergy include Microsoft-CW’s Carrie Diaries, Old Spice-FOX’s X Factor, and Elizabeth
Taylor-FOX’s Simpsons. Future research can explore potential explanations for these high and
low brand-program synergy pairs, which may be driven by the characteristics of the viewers
(e.g., demographics) and/or characteristics of the program content (e.g., product placements).
Using our estimates of αb. and γp., we can explore if brands that advertise in programs
with more socially engaged audiences also experience more online brand WOM. While
numerous studies on media context effects have found that more consumer engagement with a
television program relates positively to advertising response (e.g., Murry et al. 1992; Feltham
and Arnold 1994; Tavassoli et al. 1995; Wang and Calder 2009), other investigations have
argued that having an audience engaged with a program may actually hinder advertising
response (e.g., Danaher and Green 1997).
Figure 6 shows the posterior mean estimates of the program-specific effects on online
WOM about the program (γp,2) and about the brand (γp,1) following advertisements. We find that
twenty-five programs in our sample (30%) have positive posterior mean estimates for both γp,1
and γp,2 (upper right quadrant), indicating that these programs experience more WOM for the
program and for the brands that advertise in the program than one would expect based on other
model variables. Some examples of such programs include ABC’s Scandal, CBS’s How I Met
Your Mother, CW’s Vampire Diaries, FOX’s Glee, and NBC’s Revolution. We also observe
eleven programs (13% of our sample) that experience lower than expected online WOM for the
program and higher than expected online WOM for brands (upper left quadrant). Some examples
26
of such programs include ABC’s Revenge, CBS’s 2 Broke Girls, and FOX’s Family Guy. While
the current dominant focus of the social TV industry is on program engagement (e.g., Dumenco
2012; Lomas 2012), this result suggests that online program chatter is only part of the story and
that brands can experience increases in online WOM (all else being equal) even when they
advertised in programs with low online engagement.
[Insert Figure 6 about Here]
We also find that twenty programs in our data (24%) have higher than expected online
program chatter but that online WOM for brands is less than expected all else being equal (lower
right quadrant). These programs may be less attractive for advertisers seeking to amplify their
audience through social media. Audiences that engage online about these programs may be doing
so at the expense of online brand engagement. This finding illustrates that programs that receive
the most online WOM aren’t necessarily the best programs for advertisers. Some examples of
such programs include ABC’s Back in the Game, ABC’s Modern Family, CBS’s NCIS, FOX’s
Bones, and NBC’s Blacklist. Finally, twenty-eight programs in our data (33%) have negative
posterior means for both γp,1 and γp,2, indicating that these programs experience lower than
expected online WOM for both the program and the brands that advertise in the program given
the other model controls.
Overall, marketers interested in increasing online WOM following their television
advertising, all else being equal, may want to focus their advertising buys on programs that fall
into the top two quadrants of Figure 6 and avoid programs that fall in the bottom half of the
figure. Additionally, the results suggests that how engaged a program’s audience is online does
not necessarily indicate how the audience will engage with the advertised brands online. Some
programs that focus on being social shows online may miss the mark for advertisers while other
27
programs with low engagement may offer brands higher than expected online WOM. These
results highlight the importance of considering advertisement engagement in assessments of
social TV behavior as program engagement does not tell the whole story.
[Insert Figure 7 about Here]
In Figure 7, we investigate the relationship between program ratings and program-
specific effects on online brand WOM (γp,1). Program ratings currently dictate the cost
advertisers pay to air an advertisement during a program, with higher ratings equating to higher
costs. Marketers looking to capitalize on increased online brand chatter following their
advertisements without paying for higher ratings could look to advertise in programs that fall in
the upper-left quadrant of Figure 7. Brands that advertise in these programs, which have below
average ratings, experience higher than expected online WOM. Some examples of such
programs include CBS’s Hostages, CW’s Supernatural, FOX’s Glee, and FOX’s X Factor.
Lower ratings could be offset by higher online WOM, making these programs potential bargains
that may appeal to advertisers given that advertising costs are mostly driven by ratings.
Television networks and broadcasters could also leverage the findings in Figure 7 to negotiate
higher advertising rates in programs that offer increased online engagement for the brands that
advertise in these programs.
6. Conclusion
While the positive effects of online WOM on marketing outcomes have been widely discussed in
the literature, research on the drivers of online WOM is still in its infancy. In this investigation,
we examine if television advertising can contribute to increased online WOM for the brand
advertised and for the program in which the ad airs. Overall, our results suggest that television
28
advertising can impact online WOM for both the brand and program and that the media context
in which an advertisement airs (i.e., the television program) does influence the relationship
between television advertising and online WOM.
Our findings illustrate that social TV activity can be beneficial to marketers as it can
result in increases to online brand WOM following advertisements. Additionally, we find that
programs that receive the most online WOM aren’t necessarily the best programs for advertisers.
These are meaningful results because, given the current focus of the social TV industry on online
program engagement (e.g., Dumenco 2012; Lomas 2012), our findings suggest that advertisers
may misjudge the benefits of social TV for brands. Our results point to the need for social TV
activity to be viewed in terms of viewer engagement with both programs and advertisements.
Additionally, we find that online program WOM increases substantially following the
first advertisement in a commercial break. While increasing program engagement, this may hurt
consumer attention to advertisements airing early in the commercial break. This is relevant to
media buying strategies, as the first ad slot in a commercial break is considered to be the most
coveted ad position by advertisers (Katz 2013; Wilbur et al. 2013). Consumers’ multi-screen
behavior reveals a potential downside for advertisements airing in the first ad slot. However, we
also do find evidence that advertisers can increase online WOM for their brands following
advertisements airing in the first ad slot by incorporating digital calls-to-action, specifically a
hashtag or web address, into the creative. Thus, these results have implications for not only
advertisement design strategies but also for the importance that ad position should play in media
buy negotiations for advertisers interested in online WOM.
Our results concerning brand-program synergy further shed light on how online WOM
may be considered in the media buying process. Specifically, we find that increases in online
29
WOM from television advertising depend on brand-program synergy, with online WOM
increasing more for both brands and programs with high synergy. This finding provides an
incentive for both brands and programs interested in online WOM to consider brand-program
synergies in the media buying process. Networks may consider offering more program-specific
recommendations for marketers interested in finding the right balance between engaged program
audiences or program ratings and increases in online brand chatter.
Our analysis is subject to limitations which may be fruitful avenues of future research.
First, we do not observe if a television viewer actively sees an advertisement and subsequently
engages in online WOM. Such individual- or device-level data could allow us to attribute
changes in online WOM directly to television advertising. The narrow time window used in our
analysis helps us avoid potential unobservable influences of this limitation. Second, we focus on
the volume of online mentions. Analysis of the content of the online WOM could permit a more
nuanced investigation of the relationship between television advertising and different types of
online WOM (e.g., recommendation versus attribute-oriented WOM investigated by Gopinath et
al. 2014). Finally, while our investigation of social TV activity has focused on television
advertising’s role as a driver of online WOM, future research may build upon this foundation to
explore the relationship between television advertising, social TV, and other brand performance
measures, such as sales or brand health.
30
References
Berger, J. and E. M. Schwartz (2011), “What Drives Immediate and Ongoing Word-of-Mouth?,”
Journal of Marketing Research, 48(October), 869-880.
Berger, J. and K. L. Milkman (2012), “What Makes Online Content Viral?,” Journal of
Marketing Research, 49(April), 192-205.
Bradlow, E. T. and D. C. Schmittlein (2000), “The Little Engines that Could: Modeling the
Performance of World Wide Web Search Engines,” Marketing Science, 19(1), 43-62.
Chevalier, J. A. and D. Mayzlin (2006), “The Effect of Word of Mouth on Sales: Online Book
Reviews,” Journal of Marketing Research, 43(August), 345-354.
Copeland, C. (2013), “These are the Biggest Hurdles Facing Social TV: 2013 Could Be the
Breakthrough Year,” Adweek, (May 13), (accessed January 12, 2015), [available at
http://www.adweek.com/news/television/these-are-biggest-hurdles-facing-social-tv-149343].
Danaher, P. J. and B. J. Green (1997), “A Comparison of Media Factors that Influence the
Effectiveness of Direct Response Television Advertising,” Journal of Direct Marketing,
11(2), 46-58.
Dumenco, S. (2012), “Believe the Hype? Four Things Social TV Can Actually Do,” Ad Age,
(April 13), (accessed November 18, 2014), [available at http://adage.com/article/the-media-
guy/hype-things-social-tv/234134/].
Dumenco, S. (2013), “Q&A: The (Real) State of Social TV Right Now,” Ad Age, (April 11),
(accessed November 18, 2014), [available at http://adage.com/article/trending-topics/q-a-
real-state-social-tv/240799/].
eMarketer (2013), “US Total Media Ad Spend Inches Up, Pushed by Digital,” (August 22),
(accessed November 18, 2014), [available at http://www.emarketer.com/Article/US-Total-
Media-Advertisement-Spend-Inches-Up-Pushed-by-Digital/1010154].
Feltham, T. S. and S. J. Arnold (1994), “Program Involvement and Ad/Program Consistency as
Moderators of Program Context Effects,” Journal of Consumer Psychology, 3(1), 51-77.
Godes, D. and D. Mayzlin (2004), “Using Online Conversations to Study Word-of-Mouth
Communication,” Marketing Science, 23(4), 545-560.
Godes, D. and J. C. Silva (2012), “Sequential and Temporal Dynamics of Online Opinion,”
Marketing Science, 31(3), 448-473.
Gong, S., J. Zhang, P. Zhao, and X. Jiang (2014), “Tweets and Sales,” working paper.
Gopinath, S., J. S. Thomas, and L. Krishnamurthi (2014), “Investigating the Relationship
between the Content of Online Word of Mouth, Advertising, and Brand Performance,”
Marketing Science, 33(2), 241-258.
Hare, B. (2012), “Twice as Much TV? How Networks are Adapting to the Second Screen,”
CNN, (September 15), (accessed January 28, 2015), [available at http://www.cnn.com/2012/
09/15/showbiz/tv/second-screen-tv-our-mobile-society/].
Hill, S., A. Nalavade, and A. Benton (2012), “Social TV: Real-Time Social Media Response to
TV Advertising,” working paper.
Hoffman, D. L. and M. Fodor (2010), “Can You Measure the ROI of Your Social Media
Marketing?,” Sloan Management Review, 52(1), 41-49.
Joo, M., K. C. Wilbur, B. Cowgill, and Y. Zhue (2014), “Television Advertising and Online
Search,” Management Science, 60(1), 56-73.
Kantar Media (2014), “Kantar Media Study Confirms Link between Twitter Activity and TV
Viewing,” (September 24), (accessed September 24, 2014), [available at
31
http://www.kantarmedia.co.uk/news-views/our-latest-news/kantar-media-study-confirms-
link-between-twitter-activity-and-tv-viewing/].
Katz, H. (2013), The Media Handbook: A Complete Guide to Advertising Media Selection,
Planning, Research, and Buying. New York: Routledge.
Kumar, V., V. Bhaskaran, R. Mirchandani, and M. Shah (2013), “Creating a Measurable Social
Media Marketing Strategy: Increasing the Value and ROI of Intangibles and Tangibles for
Hokey Pokey,” Marketing Science, 32(2), 194-212.
Liaukonyte, J., T. Teixeira, and K. C. Wilbur (2015), “Television Advertising and Online
Shopping,” Marketing Science, forthcoming.
Lomas, N. (2012), “Report: Social TV Market to Be Worth $256.44BN by 2017; Europe Taking
Largest Share Now,” TechCrunch, (October 12), (accessed November 18, 2014), [available
at http://techcrunch.com/2012/10/12/report-social-tv-market-to-be-worth-256-44bn-by-2017-
europe-taking-largest-share-now/].
Manchanda, P., P.E. Rossi, and P.K. Chintagunta (2004), “Response Modeling with Nonramdon
Marketing-Mix Variables,” Journal of Marketing Research, 41(4), 467-478.
Midha, A. (2014), “Study: Exposure to TV Tweets Drives Consumers to Take Action – Both on
and off Twitter,” Twitter Media Blog, (May 8), (accessed November 18, 2014), [available at
https://blog.twitter.com/2014/study-exposure-to-tv-tweets-drives-consumers-to-take-action-
both-on-and-off-of-twitter].
Moe, W. W. and D. A. Schweidel (2012), “Online Product Opinions: Incidence, Evaluation, and
Evolution,” Marketing Science, 31(3), 372-386.
Moe, W. W. and M. Trusov (2011), “The Value of Social Dynamics in Online Product Ratings
Forums,” Journal of Marketing Research, 48(June), 444-456.
Murry, J. P., Jr., J. L. Lastovicka, and S. N. Singh (1992), “Feeling and Liking Responses to
Television Programs: An Examination of Two Explanations for Media-Context Effects,”
Journal of Consumer Research, 18(March), 441-451.
Nielsen (2012), “How Connectivity Influences Global Shopping,” (August 28), (accessed
October 15, 2014), [available at http://www.nielsen.com/us/en/insights/news/2012/how-
connectivity-influences-global-shopping.html].
Nielsen (2013), “Commercial Breaks Aren’t Tweet Breaks,” (September 18), (accessed October
15, 2014), [available at http://www.nielsen.com/us/en/insights/news/2013/commercial-
breaks-arent-tweet-breaks.html].
Nielsen (2014), “What’s Empowering the New Digital Consumer?,” (February 10), (accessed
November 18, 2014), [available at http://www.nielsen.com/content/corporate/us/en/insights
/news/2014/whats-empowering-the-new-digital-consumer.html].
Onishi, H. and P. Manchanda (2012), “Marketing Activity, Blogging and Sales,” International
Journal of Research in Marketing, 29(3), 221-234.
Poggi, J. (2012), “The Two Biggest Hurdles Social TV Still Needs to Overcome, Ad Age,
(October 4), (accessed January 12, 2015), [available at http://adage.com/article/media/
biggest-hurdles-social-tv-overcome/237576/].
Rishika, R., A. Kumar, R. Janakiraman, and R. Bezawada (2013), “The Effect of Customers’
Social Media Participation on Customer Visit Frequency and Profitability: An Empirical
Investigation,” Information Systems Research, 24(1), 108-127.
Schweidel, D. A., N. Z. Foutz, and R. J. Tanner (2014), “Synergy or Interference: The Effect of
Product Placement on Commercial Break Audience Decline,” Marketing Science, 33(6), 763-
780.
32
Schreiner, T. (2013), “Amplifiers Study: The Twitter Users Who Are Most Likely to Retweet
and How to Engage Them,” Twitter Advertising Blog, (January 10), (accessed October 15,
2014), [available at https://blog.twitter.com/2013/amplifiers-study-the-twitter-users-who-are-
most-likely-to-retweet-and-how-to-engage-them].
Stephen, A. T. and J. Galak (2012), “The Effects of Traditional and Social Earned Media on
Sales: A Study of Microlending Marketplace,” Journal of Marketing Research, 49(October),
624-639.
Subramanyam, R. (2011), “The Relationship between Social Media Buzz and TV Ratings,”
Nielsen, (October 6), (accessed October 15, 2014), [available http://www.nielsen.com/us/en/
insights/news/2011/the-relationship-between-social-media-buzz-and-tv-ratings.html].
Tavassoli, N. T., C. J. Shultz, II, and G. J. Fitzsimons (1995), “Program Involvement: Are
Moderate Levels Best for Ad Memory and Attitude Toward the Ad?,” Journal of Advertising
Research, 35(5), 61-72.
Teixeira, T., M. Wedel, and R. Pieters (2012), “Emotion-Induced Engagement in Internet Video
Ads,” Journal of Marketing Research, 49(2), 144-159.
Tellis, G. J., R. K. Chandy, and P. Thaivanich (2000), “Which Ad Works, When, Where, and
How Often? Modeling the Effects of Direct Television Advertising,” Journal of Marketing
Research, 37(1), 32-46.
Trusov, M., R. E. Bucklin, and K. Pauwels (2009), “Effects of Word-of-Mouth Versus
Traditional Marketing: Findings from an Internet Social Networking Site,” Journal of
Marketing, 73 (September), 90-102.
Wang, J. and B. J. Calder (2009), “Media Engagement and Advertising: Transportation,
Matching, Transference, and Intrusion,” Journal of Consumer Psychology, 19(3), 546-555.
Wilbur, K. C., L. Xu, and D. Kempe (2013), “Correcting Audience Externalities in Television
Advertising,” Marketing Science, 32(6), 892-912.
33
Table 1: Descriptive Statistics for Brand, Advertisement, and Program Characteristics
Group Variable Description Frequency Mean (SD)
Brand
Characteristics No Twitter
% of brands that do not have a Twitter
account 18.40
Advertisement
Characteristics
Ad break
position Ad break position in a given program 3.30 (2.21)
Ad length Ad length in seconds
25.26 (8.14)
% of ads that are less than 30 seconds 33.90
% of ads that are more than 30 seconds 1.18
Ad position in
a commercial
break
% of ads that are the first non-promo ad
in a given commercial break 17.83
% of ads that are the last non-promo ad
in a given commercial break 1.57
Calls-to-
action
% of ads that contained a phone number 6.22
% of ads that contained a hashtag 17.18
% of ads that contained a Facebook icon
or URL to a Facebook page 5.00
% of ads that contained a web address 59.19
Half-hour
break
% of ads that aired within 2 minutes of a
half-hour break 12.94
Simultaneous
ad
% of ads that aired within 2 minutes of
another ad by the same brand 7.88
Program
Characteristics
Genre % of ads on Drama/Adventure programs 46.69
% of ads on News/Magazine programs 7.37
% of ads on Suspense/Mystery/Police
programs 3.35
% of ads on Comedy programs 18.86
% of ads on Slice-of-Life/Reality
programs 23.73
Fall finale % of ads that aired during fall finale
episodes 10.49
Network % of ads on ABC 24.33
% of ads on CBS 23.97
% of ads on CW 14.35
% of ads on FOX 18.68
% of ads on NBC 18.68
Program
length Length of program in minutes
62.13 (25.11)
Program
ratings
Nielsen program ratings - reflects % of
population of TVs tuned to a particular
program for 18-49 demographic
1.85 (0.97)
Season
premiere
% of ads that aired during season
premieres 9.38
34
Table 2: Model Comparison
Model Description What's Included DIC Brand MAE Program MAE
Model 1 Baseline μbrand, μprog, Xi. 96721 9.98 145.2
(9.93, 10.02) (144.9, 145.5)
Model 2 Brand effects only μbrand, μprog, Xi., αb,1, αb,2 96629 9.88 145.1
(9.82, 9.94) (144.8, 145.4)
Model 3 Program effects only μbrand, μprog, Xi., γp,1, γp,2 96568 9.96 144.6
(9.92, 10.01) (144.1, 145.0)
Model 4 No cross effects μbrand, μprog, Xi., αb,1, γp,2 96465 9.88 144.6
(9.82, 9.94) (144.1, 145.0)
Model 5 Brand cross effects only μbrand, μprog, Xi., αb,1, αb,2, γp,2 96451 9.88 144.5
(9.82, 9.94) (144.0, 144.9)
Model 6 Program cross effects
only μbrand, μprog, Xi., αb,1, γp,1, γp,2 96533
9.85 144.6
(9.79, 9.91) (144.1, 145.0)
Model 7 Main model without
brand-program synergy μbrand, μprog, Xi., αb,1, αb,2, γp,1, γp,2 96410
9.85 144.5
(9.79, 9.92) (144.0, 144.9)
Model 8 Main model Model 7 + BPSynergybp 96201 9.82 144.6
(9.74, 9.90) (143.9, 145.7)
Note: MAE estimates are presented with 95% highest posterior density (HPD) intervals from 20,000 draws of three
independent MCMC chains. The MAE estimates are in terms of the untransformed differences between PostBWOMi
(PostPWOMi) and PreBWOMi (PrePWOMi).
35
Table 3: Impact of Brand and Advertisement Characteristics on Online WOM following Television Advertising
Brand WOM Program WOM
Brand WOM Program WOM
Variable Posterior Mean Posterior Mean
Variable Posterior Mean Posterior Mean
μbrand/μprog -0.19
(-0.65, 0.26) -2.66 ** (-3.76, -1.55)
Length - Less than
30 Seconds -0.25 ** (-0.35, -0.15) 0.11
(-0.09, 0.31)
Ad break position -0.04 * (-0.09, 0.01) 0.39 ** (0.27, 0.51)
Length - More than
30 Seconds 0.41 ** (0.10, 0.72) 0.10
(-0.57, 0.78)
Ad break position2 0.00
(-0.00, 0.01) -0.03 ** (-0.04, -0.02)
No Twitter -0.03
(-0.18, 0.13) -0.00
(-0.28, 0.27)
Ad near half-hour -0.04
(-0.13, 0.06) -0.17
(-0.39, 0.05)
Simultaneous ad 0.04
(-0.09, 0.16) 0.04
(-0.24, 0.31)
First ad -0.13
(-0.28, 0.03) 3.16 ** (2.83, 3.50)
Category
Last ad 0.03
(-0.58, 0.65) -1.24 * (-2.54, 0.05)
Apparel 0.19
(-0.21, 0.59) 0.36
(-0.40, 1.11)
Calls-to-Action
Beauty -0.04
(-0.32, 0.25) 0.39
(-0.15, 0.93)
Phone -0.23 ** (-0.43, -0.04) 0.10
(-0.30, 0.50)
Cable providers -0.12
(-0.60, 0.37) -1.32 ** (-2.18, -0.45)
Hashtag -0.02
(-0.16, 0.12) -0.12
(-0.40, 0.15)
Computer
accessories -0.31
(-0.71, 0.09) 0.16
(-0.60, 0.93)
Facebook 0.05
(-0.15, 0.26) 0.10
(-0.32, 0.51)
Computers,
notebooks, or
tablets
-0.28 * (-0.60, 0.03) 0.35
(-0.23, 0.93)
Web -0.03
(-0.13, 0.07) -0.05
(-0.25, 0.15)
Dental care -0.08
(-0.45, 0.29) 0.55
(-0.13, 1.25)
Phone X First ad -0.03
(-0.37, 0.32) -0.24
(-1.01, 0.52)
Financial -0.21
(-0.53, 0.09) 0.16
(-0.41, 0.74)
Hashtag X First ad 0.36 ** (0.14, 0.58) -0.06
(-0.54, 0.43)
Non-profit/PSAs -0.07
(-0.43, 0.30) -0.91 ** (-1.63, -0.19)
Facebook X First ad -0.05
(-0.44, 0.34) -0.66
(-1.52, 0.21)
Hair care 0.01
(-0.35, 0.37) 0.24
(-0.46, 0.94)
Web X First ad 0.17 * (-0.01, 0.35) 0.02
(-0.37, 0.42)
Insurance 0.15
(-0.18, 0.48) 0.08
(-0.56, 0.71)
Phone X Last ad 1.05
(-0.48, 2.56) -0.62
(-3.67, 2.42)
Medicine/vitamins 0.03
(-0.26, 0.32) -0.16
(-0.73, 0.41)
Hashtag X Last ad 0.10
(-0.73, 0.95) 0.11
(-1.65, 1.88)
Movies 1.49 ** (1.21, 1.77) 0.71 ** (0.17, 1.26)
Facebook X Last ad -0.65
(-2.49, 1.18) 2.14
(-1.36, 5.64)
Phones -0.30 * (-0.64, 0.03) 0.35
(-0.27, 0.95)
Web X Last ad -0.06 (-0.74, 0.61) 0.69 (-0.73, 2.12)
Wireless providers -0.05 (-0.41, 0.30) 0.46 (-0.15, 1.07)
Note: Table 3 presents posterior mean estimates along with the 95% HPD intervals from the 20,000 draws of the three independent MCMC chains.
We denote posterior mean estimates for which the 95% HPD interval excludes zero with a double asterisk (**) and estimates for which the 90%
HPD interval excludes zero with a single asterisk (*). The baseline for the category variable is other category.
36
Table 4: Impact of Program Characteristics and Brand-Program Synergy on Online WOM
following Television Advertising
Brand WOM Program WOM
Variable Posterior Mean Posterior Mean
Day program airs (Baseline: Monday)
Tuesday 0.06
(-0.10, 0.21) 0.64 ** (0.25, 1.04)
Wednesday 0.07
(-0.11, 0.26) -0.17
(-0.64, 0.31)
Thursday 0.12
(-0.06, 0.29) 0.31
(-0.13, 0.78)
Friday 0.08
(-0.11, 0.28) 0.19
(-0.29, 0.69)
Saturday 0.38
(-0.25, 0.99) 0.23
(-1.38, 1.90)
Sunday 0.16
(-0.05, 0.37) 0.58 ** (0.03, 1.14)
Fall finale -0.10 * (-0.21, 0.02) 0.30 ** (0.05, 0.54)
Genre (Baseline: Slice-of-Life/Reality)
Drama/Adventure 0.08
(-0.11, 0.27) 1.14 ** (0.60, 1.66)
News/Magazine -0.11
(-0.48, 0.27) 1.25 ** (0.19, 2.27)
Suspense/Mystery/Police 0.22
(-0.14, 0.58) 1.05 ** (0.10, 1.99)
Comedy -0.19
(-0.42, 0.05) 0.64 ** (0.02, 1.26)
Network (Baseline: ABC)
CBS -0.12
(-0.28, 0.05) 0.29
(-0.16, 0.75)
CW 0.22 * (-0.01, 0.45) 0.53 * (-0.08, 1.15)
FOX 0.14
(-0.04, 0.33) -0.95 ** (-1.47, -0.45)
NBC -0.05
(-0.23, 0.14) -0.54 ** (-1.06, -0.03)
Program length -0.00
(-0.00, 0.00) -0.00
(-0.01, 0.00)
Program rating 0.23 ** (0.16, 0.31) -0.16 * (-0.33, 0.02)
Season premiere 0.09
(-0.03, 0.21) 1.02 ** (0.76, 1.29)
Time program airs (Baseline: 8:00pm)
8:30pm 0.07
(-0.06, 0.20) -0.49 ** (-0.79, -0.19)
9:00pm 0.10
(-0.04, 0.24) -0.24
(-0.57, 0.08)
9:30pm 0.08
(-0.08, 0.24) -0.50 ** (-0.89, -0.11)
10:00pm 0.02
(-0.16, 0.20) -0.32
(-0.77, 0.12)
10:30pm 0.13
(-0.07, 0.34) 0.01
(-0.49, 0.50)
Brand-program synergy (BPSynergybp) 0.08 ** (0.02, 0.14) 0.15 ** (0.05, 0.27)
Note: Table 4 presents posterior mean estimates along with the 95% HPD intervals from the 20,000 draws
of the three independent MCMC chains. We denote posterior mean estimates for which the 95% HPD
interval excludes zero with a double asterisk (**) and estimates for which the 90% HPD interval excludes
zero with a single asterisk (*).
37
Figure 1: Change in Online Brand WOM following Television Advertisements across Brands in Different Categories
38
Figure 2: Change in Online Brand WOM following Television Advertisements across Ad and Program Characteristics
Figure 3: Change in Online Program WOM following Television Advertisements
39
Figure 4: Change in Online Brand Mentions from Television Advertisements Airing on ABC’s 20/20, FOX’s Glee, and ABC’s
Scandal
Figure 5: Change in Online Brand WOM following Television Advertisements for Sprint and Gravity across Programs
September 23, 2013 Primetime Online Twitter Conversations about Sprint
September 23, 2013 Primetime Online Twitter Conversations about Gravity
40
Figure 6: Relationship between Program-specific Effects on Online Program WOM (γp,2) and Online Brand WOM (γp,1)
following Television Advertising
41
Figure 7: Relationship between Average Program Ratings and Program-specific Effects on Online Brand WOM (γp,1)
following Television Advertising
Note: The quadrant lines in Figure 7 are drawn at zero for γp,1 and the overall mean program rating in our sample (1.85).
42
Online Appendix
This Online Appendix presents summary statistics and empirical results from the main model as
well as empirical results from several alternative model specifications that were excluded from
the main text for conciseness.
A.1 Data Overview
Tables A1 and A2 present a detailed overview of the 15 categories and the 84 programs explored
in our analysis. These tables show the frequency of advertising instances by category and the
most advertised brands by category and program genre. Additionally, Figure A1 shows the
distribution of ad instances by day of the week, time block, and program rating. Finally, Table
A3 presents a summary of the fifty-eight brand, advertisement, and program characteristics
discussed in the Data Description section which are included in Xi. in equation (3).
A.2 Searching Brand Mentions on Twitter
Table A4 presents the search terms used to capture brand mentions on Twitter for a subsample of
the brands used in our analysis. Recall that we capture Twitter conversations that mention the
brand, a hashtag featuring the brand name, a hashtag included in the brand’s advertisement, or
the brand’s Twitter handle. Parent brand names were used as the focus of the search over product
brand names (e.g., “Colgate” versus “Colgate Optic White”) for almost all brands with the
exceptions being motion pictures, books, and tech products. For these exceptions, product brand
names were incorporated into the search of Twitter conversations for brand mentions.
Additionally, in some cases a brand shares a name with a common word, such as Halls or
Nationwide. For such brands, we also make use of product brand names to focus the search of
Twitter conversation on brand chatter. Examples can be seen in Table A5.
43
A.3 Additional Results from Equations (1)-(6)
Table A6 shows the posterior mean estimates of the variance and heterogeneity parameters from
equations (1)-(3). Table A7 shows the results from the model of advertisers’ choices of programs
shown in equations (4)-(6).
A.4 Alternative Model Specifications
We estimate several alternative specifications of the joint model of the change in online WOM
for brands and programs following television advertising. First, we estimate our modeling
framework using dependent variables that are not constructed as difference measures. Next, we
evaluate our model when negative online brand and program mentions are excluded from the
data. Finally, we estimate sixteen alternative specifications utilizing different pre- and post-
advertisement time windows.
Non-Difference Dependent Variables. We estimate our modeling framework in which the
dependent variables are not constructed as difference measures but, rather, are specified as
follows:
(A1)
,
ˆ
ˆ~
2
1
i
i
i
i
Y
YN
PostPWOM
PostBWOM
(A2)
58
1 2
1][],[
2
1
2
1
2],[
1],[
2],[
1],[
2
1
ˆ
ˆ
k
ikk
kipib
i
i
ip
ip
ib
ib
prog
brand
i
i XBPSynergyPrePWOM
PreBWOM
Y
Y
where PostBWOMi (PostPWOMi) is the number of online Twitter mentions for brand b (program
p) two minutes after b’s advertisement airs in p and PreBWOMi (PrePWOMi) is the number of
online mentions for brand b (program p) two minutes before b’s advertisement airs in p. We
jointly estimate equations (A1) and (A2) along with the model of the advertisers’ program
44
choices specified in equations (4)-(6) using the same estimation procedures detailed in Model
Development.
The substantive results from this alternative specification are consistent with those from
our main model estimation. We highlight a few key similarities and discuss two notable
differences below. To begin, the posterior mean estimates for φ. are positive (95% highest
posterior density (HPD) intervals exclude zero). Additionally, as found in our main analysis,
brand-program synergy (BPSynergybp) positively impacts online WOM for both brands and
programs. This result indicates that if brand b and program p have high synergy, both will
experience increases in online WOM (all else being equal) following b’s advertisement airing in
p. This further illustrates the robustness of the result that the media context in which a television
advertisement airs influences the ad’s impact on online WOM.
There are two key differences to note between the estimation results for the alternative
specification detailed in equations (A1) and (A2) and our main model estimation. First, the 90%
HPD interval for the interaction of featuring a web address in an advertisement and the first ad
position does not exclude zero in this non-difference dependent variable specification. Thus, in
this alternative specification, we find that only one digital call-to-action, hashtags, increases
online brand WOM after airing in the first ad slot of a commercial break. Second, given the
explanatory power of PrePWOMi, many of the control variables in Xi. no longer have significant
impacts on online program chatter in the program WOM model.
Excluding Negative WOM. We also estimate an alternative model in which we consider
the valence of the online conversations. In our primary analysis, we do not distinguish between
positive, neutral, and negative WOM since we are interested in the volume of conversations and
understanding how television advertising drives online WOM overall. However, we recognize
45
that some marketers and television networks may be interested in how television advertising
drives positive and neutral online WOM about brands and programs. To explore this
relationship, we leverage the valence coding of the online brand and program mentions provided
by Topsy Pro. Topsy uses a proprietary software to code Tweets as positive, neutral, or negative
by analyzing the weighted sentiment of words and phrases using an automated process which
they validate through manual checks of Tweet content. For this alternative model, we exclude
from the data brand and program Twitter mentions that Topsy codes as negative and use the
same modeling framework and estimation procedure discussed in Model Development.
The substantive results between this alternative specification and our main model
specification are the same. In general, the estimation results are nearly identical between these
two specifications. For example, only four (3.4%) of the posterior mean estimates for the 118
coefficients in η. and θk. change sign between these two specifications, and the 90% HPD
intervals for all four of these estimates include zero in both specifications. The only notable
difference between these two specifications is that only the 90% HPD interval for η1 excludes
zero in this alternative specification, while in the main model specification, the 95% HPD
interval for η1 excludes zero.
Different Time Windows. We estimate a series of alternative models with different pre-
and post-advertisement time windows to test the robustness of our results as time from the
advertisements’ airings increases. The modeling framework and estimation is the same as the
main model; only the length of PostBWOMi, PreBWOMi, PostPWOMi, and PrePWOMi is
changed. We explore the following time lengths for these four variables for a total of sixteen
alternative specifications: 3-15 minutes, 30 minutes, 45 minutes, and 60 minutes. Recall that at
the most granular level, advertisers can aim to air their advertisements in specific quarter-hour
46
periods (Liaukonyte et al. 2015). Thus, for the alternative models in which PostBWOMi,
PreBWOMi, PostPWOMi, and PrePWOMi are eight minutes or longer (such that the total time
window would exceed fifteen minutes), it would be feasible for advertisers or television
networks to choose certain time windows to air advertisements in order to impact online WOM.
While feasible, we argue that this is very unlikely. To elaborate, 80% of the advertiser-network
negotiations are completed in the May up-front market several months prior to the start of the fall
television season, and timing of the ad is not stipulated in advertiser-network contracts
(Liaukonyte et al. 2015). Furthermore, television networks commonly assign advertisements to
commercial breaks at random (Wilbur et al. 2013). Thus, even as we explore the impact of
television advertising on online WOM as the time since the advertisement’s airing increases, we
do not think it is probable that the results from these alternative models suffer from the
endogeneity concern that advertisers or television networks could choose a certain time window
to air an advertisement in order to impact online WOM.
The results across these sixteen alternative specifications are very consistent, and we see
significant evidence that are substantive findings concerning the relationship between television
advertising and online WOM hold as time since the advertisements’ airings increases. In fourteen
of the sixteen alternative models (88%), we find that brand-program synergy (BPSynergybp)
impacts online WOM following television advertising. These effects are most persistent on
online chatter about programs, as BPSynergybp continues to have a positive influence (95% HPD
interval excludes zero) on online program WOM even sixty minutes after the advertisement airs
(compared to sixty minutes before). Additionally, in fourteen of the sixteen alternative
specifications (88%), we also see evidence that using digital calls-to-action in an advertisement
that airs in the first ad slot influences online brand WOM. The substantial impact of featuring
47
hashtags in the first ad of a commercial break continues to have a positive impact on online
brand chatter even sixty minutes after the advertisement airs (compared to sixty minutes before).
We additionally see other effects of utilizing calls-to-action in television advertising on
online WOM emerge as time since the advertisements’ airings increases. For example, as time
since the advertisements’ airings increases, we find that featuring a phone number in a television
advertisement no longer negatively impacts online brand WOM and that featuring a Facebook
page link or icon in an advertisement that airs first in a commercial break negatively impacts
online WOM about the brand on Twitter (95% HPD intervals for the posterior mean estimate
exclude zero). Additionally, we find that television advertisements featuring web addresses that
air in the last ad slot in a commercial break negatively impact online program WOM (95% HPD
interval for the posterior mean estimate excludes zero) as the time since the advertisement’s
airing increases.
48
References
Liaukonyte, J., T. Teixeira, and K. C. Wilbur (2015), “Television Advertising and Online
Shopping,” Marketing Science, forthcoming.
Wilbur, K. C., L. Xu, and D. Kempe (2013), “Correcting Audience Externalities in Television
Advertising,” Marketing Science, 32(6), 892-912.
49
Table A1: Advertising Instances and Most Advertised Brands by Category
Category Description of Category Advertisement
Instances (%) Most Advertised Brands by Category
Apparel Apparel, jewelry, and shoes 183 (2.01%) Cotton Incorporated, Fruit of the Loom,
Forevermark, Hanes, Skechers
Beauty Beauty care, makeup, fragrances, and
toiletries 1220 (13.40%)
Maybelline, L'Oréal, Neutrogena, Clinique/Cover
Girl (tie)
Cable providers Cable networks, satellite providers, and
internet webcasts 135 (1.48%)
WIGS (Web Channel), DirecTV, Big Ten Network,
Showtime, Fox Deportes
Computer accessories Computer accessories, components, and
software 171 (1.88%) Microsoft, Intel, Google, Apple/Canon (tie)
Computers,
notebooks, and tablets Notebook, laptop, and tablet computers 950 (10.44%) Microsoft, Amazon, Google, Apple, Lenovo
Dental care Toothpaste, toothbrushes, whiteners,
mouthwashes, and dental supplies 274 (3.01%) Crest, Colgate, Sensodyne, Listerine, Act
Financial
Non-insurance financial products, credit
cards, banking, retirement services,
investments, and loan/credit products
937 (10.29%) Bank of America, Citibank, Capital One, Chase,
American Express
Hair care Shampoos, hair color, hair styling, and hair
accessories 216 (2.37%) Garnier/Tresemme (tie), L'Oréal, Clairol, Pantene
Insurance Automobile, medical, life, and homeowners
insurance products 508 (5.58%)
Farmers Insurance, Geico, United Healthcare,
Progressive, State Farm
Medicine/vitamins Digestive aids, cold/cough/pain/sleep
remedies, and vitamins/minerals 934 (10.26%) DayQuil/NyQuil, Tylenol, Advil, Zicam, Zyrtec
Movies Motion pictures 1227 (13.48%)
Gravity, The Hobbit: The Desolation of Smaug, The
Counselor, Frozen, Captain Phillips/Last Vegas
(tie)
Non-profit/PSAs Public service announcements and non-
profit/U.S. Government announcements 243 (2.67%)
CBS Cares, FosterMore, Autism Speaks/Common
Sense Media/It Can Wait (tie)
Phones Phones 892 (9.80%) Apple, Nokia, Motorola, Samsung, LG
Wireless providers Wireless/internet telecom providers 976 (10.72%) AT&T, Sprint, Verizon, T-Mobile, Virgin Mobile
Other Products not included in the above categories 237 (2.60%) Pampers, USPS, Hallmark/NFL (tie),
Dr. Scholls/Luvs (tie)
50
Table A2: List of Programs by Program Genre
Program Genre Programs (Network) Most Advertised Brands by Genre
Drama/Adventure
Almost Human (FOX), Arrow (CW), Beauty and the Beast (CW), Betrayal
(ABC), Blacklist (NBC), Blue Bloods (CBS), Bones (FOX), Carrie Diaries
(CW), Castle (ABC), Chicago Fire (NBC), Criminal Minds (CBS), Dracula (NBC), Elementary (CBS), Glee (FOX), Good Wife (CBS), Grey's Anatomy
(ABC), Grimm (NBC), Hart of Dixie (CW), Hawaii Five-0 (CBS), Hostages
(CBS), Lucky 7 (ABC), Marvel's Agents of S.H.I.E.L.D. (ABC), Nashville
(ABC), NCIS (CBS), NCIS: Los Angeles (CBS), Once Upon a Time (ABC),
Once Upon a Time in Wonderland (ABC), Originals (CW), Parenthood (NBC), Person of Interest (CBS), Reign (CW), Revenge (ABC), Revolution
(NBC), Scandal (ABC), Sleepy Hollow (FOX), Supernatural (CW), Tomorrow
People (CW), Vampire Diaries (CW)
Apple, Microsoft, AT&T, Nokia,
Maybelline
News/Magazine 20/20 (ABC), 48 Hours (CBS), Dateline (NBC) FosterMore, Citibank, Sprint,
AT&T/Zicam (tie)
Suspense/Mystery/Police CSI (CBS), Ironside (NBC), Law & Order: SVU (NBC), Mentalist (CBS) Apple, Citibank, Google, AT&T,
Chase/Microsoft/Nokia/Sprint (tie)
Comedy
2 Broke Girls (CBS), American Dad! (FOX), Back in the Game (ABC), Big
Bang Theory (CBS), Bob's Burgers (FOX), Brooklyn Nine-Nine (FOX), Crazy Ones (CBS), Dads (FOX), Family Guy (FOX), Goldbergs (ABC), How I Met
Your Mother (CBS), Last Man Standing (ABC), Michael J. Fox Show (NBC),
Middle (ABC), Millers (CBS), Mindy Project (FOX), Modern Family (ABC),
Mom (CBS), Neighbors (ABC), New Girl (FOX), Parks and Recreation
(NBC), Raising Hope (FOX), Sean Saves the World (NBC), Simpsons (FOX),
Super Fun Night (ABC), Trophy Wife (ABC), Two and A Half Men (CBS), We
Are Men (CBS), Welcome to the Family (NBC)
Microsoft, Nokia, Apple, Verizon,
Bank of America
Slice-of-Life/Reality
Amazing Race (CBS), America's Next Top Model (CW), Biggest Loser (NBC),
Dancing with the Stars (ABC), MasterChef Junior (FOX), Shark Tank (ABC),
Survivor (CBS), Undercover Boss (CBS), Voice (NBC), X Factor (FOX)
Bank of America/Sprint (tie),
Microsoft, Nokia, Apple
51
Table A3: Brand, Advertisement, and Program Characteristics in Xik
Group Parameter Variable Description
Brand
Characteristics
Brand category Xi1-Xi14 Dummy variables for the categories listed in Table A1 (BASELINE: other
category)
No Twitter Xi15 Dummy variable for if a brand does not have a Twitter account
Advertisement
Characteristics
Calls-to-action
(CTAs) Xi16-Xi19
Dummy variables for if ad contains a hashtag, Facebook icon or URL for a
Facebook page, phone number, and/or web address
Ad length Xi20-Xi21 Dummy variables if ad is less than 30 seconds or if ad is more than 30 seconds
(BASELINE: 30 second ad)
Ad position in
commercial break Xi22-Xi23
Dummy variables if ad appears in first slot in a given commercial break or if ad
appears in last slot of a given commercial break (BASELINE: ad in a middle slot of
the commercial break)
Ad break position in
program Xi24-Xi25
Ad break position in a given program and quadratic of ad break position in a given
program
Half-hour break Xi26 Dummy variable for if ad airs within two minutes of a half-hour interval
Simultaneous ad Xi27 Dummy variable for if a brand's ad runs in a simultaneous commercial break with
another ad instance by the same brand
CTA and ad position
in break interactions Xi28-Xi35
Interactions between the four call-to-action dummy variables and the two ad
position in commercial break dummy variables
Program
Characteristics
Program rating Xi36 Nielsen program rating
Network Xi37-Xi40 Dummy variables for if the program airs on CBS, CW, FOX, or NBC (BASELINE:
ABC)
Genre Xi41-Xi44 Dummy variables for if the program is a Drama/Adventure, News/Magazine,
Suspense/Mystery/Police, or Comedy (BASELINE: Slice-of-Life/Reality program)
Special Episodes Xi45-Xi46 Dummy variables for season premiere and fall finales episodes
Program length Xi47 Length of program in minutes
Day of the week Xi48-Xi53 Dummy variables for day of the week the ad airs (BASELINE: Monday)
Time Xi54-Xi58 Dummy variables for time the ad airs created in half-hour increments from 8:00pm-
11:00pm (Baseline: 8:00pm)
52
Table A4: Examples of Terms Used to SearchTwitter Conversations about Brands
PARENT BRAND PRODUCT BRAND SEARCH TERMS
Brand Name(s) Brand Hashtag(s) Hashtag used in Ad Brand Twitter Handle
Amazon Amazon Kindle Fire
HDX: Tablet Computers "Kindle"
"#Kindle" OR
"#KindleFire" OR
"#KindleFireHDX" OR
"#AmazonKindle"
"@AmazonKindle"
Apple Apple iPad Air: Tablet
Computer "iPad"
"#ipad" OR
"#appleipad" OR
"#ipadair"
Aveeno
Aveeno Daily
Moisturizing Body
Lotion
"Aveeno" "#aveeno"
DirecTV DirecTV "DirecTV" "#DirecTV" "@DIRECTV"
Hulu Hulu Plus "Hulu" "#hulu" "@hulu"
Lions Gate Pictures The Hunger Games:
Catching Fire
"Hunger Games" OR
"Catching Fire"
"#TheHungerGames"
OR "#HungerGames" "#CatchingFire" "@TheHungerGames"
Old Spice Old Spice Toiletries "Old Spice" "#OldSpice" "@OldSpice"
Revlon Revlon Lash Potion "Revlon" "#revlon" "#CastASpell" "@revlon"
Special K Special K Protein Meal
Bar & Protein Shake "Special K" "#SpecialK" "@SpecialKUS"
T-Mobile T-Mobile Consumer
Wireless Service
"T-Mobile" OR
"TMobile" OR "T
Mobile"
"#TMobile" "#Hate2Wait" "@TMobile"
Tylenol Tylenol "Tylenol" "#Tylenol"
53
Table A5: Examples of Terms Used to Search Twitter Conversations about Brands that Share a Name with a
Common Word
PARENT BRAND PRODUCT BRAND SEARCH TERMS
Brand Name(s) Brand Hashtag(s) Hashtag used in Ad Brand Twitter Handle
Coach Coach Poppy Blossom:
Women's Fragrance
("Coach" AND
"Poppy Blossom")
"#CoachPoppyBlossom"
OR ("#Coach" AND
"Poppy Blossom")
("@Coach" AND
"Poppy Blossom")
Halls Halls Mentho-Lyptus:
Cough Remedy Tablets
("Halls" AND
"cough")
("Halls" AND
"#cough") OR ("#Halls"
AND "cough")
Luvs Luvs Disposable
Diapers
("Luvs" AND
"diapers") "#luvs" "@Luvs"
Nationwide Nationwide Auto
Insurance
("Nationwide" AND
"insurance") "#Nationwide" "@Nationwide"
Outlook Outlook.com
"Outlook.com" OR
("Outlook" AND
"email")
"#Outlook" OR
"#Outlook.com" "@Outlook"
Warner Bros
Pictures Prisoners
("Prisoners" AND
"movie") "#theprisonersmovie" "#prisonersmovie"
54
Table A6: Results of Television Advertising’s Effect on Online WOM
Brand WOM Program WOM
Variable Parameter Posterior Mean Variable Parameter Posterior Mean
Variance
for Yi1 Τ11 2.39 (2.32, 2.46)
Variance
for Yi2 Τ22 11.83 (11.49, 12.19)
Covariance Τ12 -0.002 (-0.113, 0.108) Covariance Τ21 -0.002 (-0.113, 0.108)
Heterogeneity
for αb1 τα1 0.07 (0.04, 0.11)
Heterogeneity
for αb2 τγ1 0.09 (0.03, 0.17)
Heterogeneity
for γp1 τα2 0.04 (0.02, 0.07)
Heterogeneity
for γp2 τγ2 0.37 (0.22, 0.59)
Note: Table A6 presents posterior mean estimates along with the 95% highest posterior density (HPD)
intervals from the 20,000 draws of the three independent MCMC chains.
Table A7: Results of from Model of Advertisers’ Choices of Program(s)
Variable Mean
Parameter
Posterior
Mean
Heterogeneity
Parameter
Posterior
Mean
Brand location on dimension 1 1bB 2.28
τB1 2.15
(1.58, 3.24) (1.57, 2.86)
Brand location on dimension 2 2bB 0.76
τB2 0.57
(0.22, 1.24) (0.28, 1.02)
Program location on dimension 1 1pP 2.57
τP1 0.70
(1.85, 3.58) (0.46, 1.05)
Program location on dimension 2 2pP -0.57
τP2 0.36
(-1.14, -0.08) (0.23, 0.55)
Slope 3.73
τλ 4.32
(3.29, 4.22) (3.18, 5.79)
Mean (SD)
Summary Statistics for LatentDistbp 2.20 (0.77)
Summary Statistics for
BPSynergybp 0.54 (0.30)
Note: Table A7 presents posterior mean estimates along with the 95% HPD intervals from the 20,000
draws of the three independent MCMC chains.
55
Figure A1: Distribution of Advertisement Instances by Day of the Week, Time Block, and
Program Rating
Note: The times labeled in “Distribution of Ad Instances by Time Block” are in Eastern Standard Time
(EST).
top related