Re-imagining Television Audience Research: Tracing Viewing Patterns on Twitter

Abstract

IntroductionIn his seminal article, “Communications: Blindspot of Western Marxism” (1977), Dallas Smythe suggested that audiences are the commodity form of advertiser-supported communications, as their time is sold to advertisers. Audience measurement firms establish the audience size for a programme by calculating how many people are “tuned in” to a particular offering, and then provide their estimates to advertisers and break down their figures on the basis of demographic characteristics (these characteristics include age, gender, and income level). These ratings have long been the currency of the television industry. Essentially, Smythe points out that advertisers purchase, “the services of audiences with predictable specifications who will pay attention in predictable numbers and at particular times to particular means of communication” (4). Ien Ang has proposed that audience measurement produces an “objectified category of others” that can be governed and abstracted from the “messiness of everyday life” (8, 132). Indeed, Ang sees ratings to be a means of controlling the audience by creating a truth about them that suits the industry’s needs for an exchangeable commodity.In the United States, Nielsen ratings dictate the terms for the buying and selling of television advertising. Over the years, Nielsen has adjusted the measurement methodology to satisfy the demands of various stakeholders: audience measurement companies, advertisers, programme producers, and network executives, among others. Recently, however, social media (particularly Twitter) has threatened Nielsen’s preeminence. Writing in Wired magazine in 2013, Tom Vanderbilt went so far as to declare that the Nielsen Family—the “25,000 households whose TV habits collectively provide a statistical snapshot of a nation’s viewing behavior” (n.p.)—was now dead. He proposed that a show’s “tweetability” had become more important than its Nielsen rating.Nielsen, for its part, has tried to keep up with the changing television landscape and the demands of the television industry. In 2012 they partnered with McKinsey & Company to create the social media consulting company NM Incite, and acquired social TV startup SocialGuide. The following year the company introduced Nielsen Twitter TV Ratings (NTTR) as a supplement to its traditional ratings offering. This step is in line with the shifting industry interest from measuring audience exposure to programming to measuring audience engagement with programming (Jenkins; and Napoli).With NTTR, Nielsen has made, we suggest here, a fairly unimaginative and restricted addition to existing metrics in that it limits its measurements to tweet volume and tweet impressions. In this paper we explore other ways Twitter might be used to create insights that would be useful for audience research. Richard Rogers has raised the question of whether and when standard methods should be applied to the study of a new medium (162). We respond by proposing that, in the case of NTTR, traditional methods should not be applied to Twitter.We begin by briefly discussing the emergence of social media metrics and some of the problems involved in employing these metrics in current audience research. We then investigate how Twitter invites new forms of inquiry, drawing a picture of relationships among television programmes based on viewer tweets. In this re-imagining of audience research, following the Digital Methods tradition, we treat Twitter as a “postdemographic machine” (Rogers) that profiles user tastes, interests, favourite things, and so forth (rather than demographics such as age, income, educational level, and ethnicity).Nielsen and the Introduction of NTTRNielsen collects data about television viewing through diaries kept by members of a relatively small audience sample and meters that are connected to television sets. They provide ratings for programmes according to a system where one Nielsen rating point equals one per cent of all US households with television sets tuned into that programme. Two trends now strain this traditional form of the “exposure metrics” used in the buying and selling of primetime advertising: audience fragmentation and audience autonomy (Napoli). These terms refer, respectively, to the explosion of channels and platforms, first via cable television and later the Internet, on which viewers can watch television programming, and to viewers’ increased control over what television programmes they watch and when they watch them, thanks to technologies such as remote control, DVR, and now the Internet. These trends have eroded audience size for broadcast television and have made traditional metrics, which measure a sample of the audience, increasingly less representative of the viewing population as a whole. Responding to the changing television landscape, Nielsen introduced its “C3 rating” in 2009. This rating measures commercials watched both during first-run broadcasts and on DVR playback within three days (Nielsen Company, “C3 TV Ratings”). In this new landscape, producers and advertisers have begun to think that a small, yet engaged, group of viewers might be more valuable than a larger, more superficial audience (Jenkins 63). They have become increasingly interested in viewers’ engagement with particular programmes. Since around 2009, social TV as a television strategy—to stimulate people to watch television at its scheduled broadcast time and to deepen their engagement with programmes using the real-time features of social media—has gained prominence (van Es). Social TV efforts protect the existing business model for television.The Internet, and its communication structures, are becoming a valuable companion to television, not only because social media reinvigorates first-run viewing, but also because it provides data about viewing activity (Lee and Andrejevic). Social TV’s popularity made the introduction of NTTR unsurprising. Moreover, the particular partnership with Twitter, as opposed to other social platforms, makes sense, because Twitter is—at least for now—the biggest player in the social TV space. Its current ascendency may be due to the particular public openness of the platform, which unlike Facebook allows even non-account holders access to Twitter streams, and its users’ propensity to share their responses to TV on Twitter in real time (Proulx and Shepatin 13).NTTR measures the total number of tweets that refer to a specific television episode, the number of times these tweets were viewed (“impressions”), “unique authors” (accounts that tweeted at least once about a specific episode), and “unique audience” (the number of individual accounts that received at least one “impression” of the tweets about a specific episode [Nielsen Company, “Weekly Top Ten”]). Since May 2014, Nielsen also includes a demographic breakdown in NTTR, specifying the age and gender of those who tweet and view tweets (related to programming from 250 US TV networks). Through a partnership with GfK, a leading market research institute in Europe, Nielsen has since introduced Twitter TV ratings in Germany, Austria, and The Netherlands.In the United States, other companies besides Nielsen generate social TV analytics. Philip Napoli has compared the leading three social TV analytics providers: BlueFin Labs, Trenddr.tv, and General Sentiment. Twitter has recently acquired the first two of these firms as part of its efforts to solidify its position in the social TV landscape. These social TV analytics providers, Napoli claims, and we would add NTTR to the list, are methodologically distinct from traditional ratings in three ways. First, they track everyone who is tweeting about a programme rather than using a “representative” sample. Second, people do not receive incentives to participate in the research, or even get to opt in or out of it. Third, social analytics can focus on not only the “volume” but also the “valence” of an online conversation: it can assign, for instance, a quantitative score between 1 and 10 to reflect either positive or negative contributions on social media (Napoli 11).Among the reviewed providers, Napoli found two main methodological disparities: the platforms they draw data from and the time windows used (10-15). He contends that by measuring different factors they offer different interpretations of “engagement” and give conflicting representations of the audience as a commodity. Social media metrics are not going to work as long as there is disagreement over how to measure and value television’s viewers.Social media metrics have been met with considerable criticism. Like traditional metrics, they track a particular demographic rather than a random sample of people, and so are not broadly representative. Nancy Baym points out how social media metrics in audience research are affected by factors such as “skew,” a by-product of the fact that platforms actively shape the communication that takes place on them. Trending topics on Twitter may, for instance, boost the number of tweets about a programme. She also identifies the problem of deception: bots can tweet about topics and accounts can purchase certain forms of engagement (Baym n.p.).Most important here, perhaps, is what Baym calls “ambiguous meaning”: actions on social media are “uncoupled from contexts of action and application” (Dean in Baym n.p.). In the case of Twitter, for instance, it is not readily evident why people tweet, or why they retweet or favourite certain tweets; one can learn why people do so only through methods such as interviews.The discussion of these limitations highlights the need for a certain sensibility when encountering social media metrics. The limitations themselves, however, do not mean that Twitter is ineffectual for audience research. Tweets can help generate insights and raise new questions about television viewing. Between Counting Viewers and Counting TweetsTo explore the relationship between traditional ratings and NTTR, we collected tweets about television programmes in The Netherlands during the first four weeks of September 2014. This project was conducted, on behalf of BuzzCapture, by a group of research assistants of the Utrecht Data School (Leila Essanoussi, Friso Leder, David de Wied, and Koen Mooij) under our instruction. Specifically, we extracted tweets from 1 September up to, and including, 29 September 2014. We included one extra day since programmes aired on Sunday 28 might still have been discussed around midnight. Initially, we collected tweets on the basis of the official and popular hashtags relating to the 30 most-watched television programmes (rated by the national association for audience research, Stichting KijkOnderzoek, SKO); we then added two programmes not included in this list that were frequently mentioned on Twitter. We collected tweets referring to these 32 programmes as well as profile information of the related Twitter accounts. After removing marketing and spam accounts, we had a sample of 135,882 tweets posted by 39,792 unique tweeters.Figure 1: Number of Viewers versus Average Number of TweetsWe then compared the number of viewers to the average number of tweets referring to the 32 television programmes in a scatterplot (see Figure 1). We took the average number of tweets as our reference point to correct for the fact that the frequency of broadcasting differed among the programmes. Figure 1 shows that some programmes attract a large audience but generate few tweets, and vice versa. For example, Het Journaal, with three million viewers, generates an average of 160 tweets per broadcast, while Pauw, with fewer than 750,000 viewers, generates on average nearly 1,000 tweets.This sort of disparity suggests that what is “successful” in terms of the number of tweets may not be “successful” in terms of the number of viewers. There are several possible explanations for the variation in Twitter activity: a political talk show like Pauw consists of highly controversial content, making it more likely to “spark” tweets and retweets, while the eight o’clock news airs less polarising points of view. Moreover, reality shows like The Voice of Holland not only stir up conflict and invite enthusiastic judgements (Bratich) but also actively encourage their audience to interact through social media.Our sample, moreover, suggests that viewing television and tweeting about programming constitute two distinct phenomena. However, there remains a lot of speculation about what can be inferred from a tweet and tweet impressions, and thus what price tag to attach to these sorts of activities. Twitter numbers are now used either as a point of differentiation from traditional methods (such as, to sell programmes by claiming that they are successful, despite their low ratings), or when a programme’s audience is too small to be registered by traditional methods (Napoli). In what follows, we explore how tweets can be used to study viewing patterns, and briefly consider the advantages of doing so.Looking at Affiliations among TV Programmes through Tweets In his book Digital Methods (2013), Richard Rogers points out how social networking sites allow for new methods to study social networks. Information supplied to social media platforms can be used to explore “post-demographics,” meaning that they can be used to profile users’ tastes, interests, and favourite items, and the co-occurrences of the expressions of these preferences (154). Although this approach is common on various platforms (for example, in Amazon recommendations) and in online marketing practices (as in those that establish affiliations among the brands people tweet about), it has not commonly been used to research audiences. Looking at affiliations can, we suggest here, help create new knowledge about audiences.Figure 2: The Overlap in Tweeters among 32 Programmes in the NetherlandsUsing the same dataset of tweets used for the scatterplot, we tracked the viewing patterns of tweeters, analysing the sequence in which they used programme hashtags. We found that 8,958 people tweeted about more than one programme. The data revealed very interesting results when we calculated the relative overlap among programmes, charting the number of interrelating tweeters with respect to the absolute number of tweeters who referred to the two respective programmes. We imported the 32 nodes (the programmes) and the relative relations to Gephi in order to generate an association network, using the force-directed layout algorithm ForceAtlas2. The resulting network helps illuminate which programmes attract the same tweeters (see Figure 2). Our decision to rectify for the bias of highly social programmes has serious consequences and its validity is open to discussion. We did so to help expose taste relations (rather than reflect popularity).The association network demonstrates that TV shows of the same genre attract similar Twitter audiences: Dubbeltje op Zijn Kant and Uitstel van Executie are both reality shows about personal financial struggles, Studio Sport and Studio Voetbal are sport programmes, Hart van Nederland and RTL Boulevard are tabloid news shows, and Spoorloos and Familiedinner are programmes that centre on family issues. Aside from the strong overlap between programmes of the same genre, the visualisation also shows a concentration of programmes from public broadcasters—on the left-hand side of the figure—and those on commercial television—seen on the right. These connections suggest that people that watch commercial television tend to focus their viewing to commercial television (and the same is true for public television). The Voice of Holland, which seems to have a weak overlap in tweeters with multiple programmes, presents an intriguing case. This observation invites further consideration of its audience composition (which traditional ratings might help with).These are just some quick reflections made possible by using different methods to study Twitter. Although the input from an association network does not provide neat numbers that can serve as a “commodity,” it could help inform the programme schedules of television networks (they could adjust air times to better fit audience preferences, for example, by scheduling two TV shows with similar Twitter audiences in back-to-back time slots). Such insights could assist advertisers better understand consumer behaviour and viewing habits and thus maximise the effectiveness of their commercials. Television producers could also explore on-air and online collaborations between programmes. ConclusionIn this paper we have discussed the limitations of both traditional metrics and newer social media metrics. We explored how tweets can be used to generate insights into viewing patterns, briefly considering how such findings could benefit various parties. We have shown that the counting of tweets addresses the tweetability of a show but seems unrelated to the show’s number of viewers. We speculate, also, that programmes that spark polarised debate or motivate users to engage through social media are receiving many more mentions on Twitter than other sorts of programming. There is much space for TV programmers to build new relationships with their viewers.We have offered some criticism on the decision of NTTR to apply old methods to a new medium, and proposed that audience research on social media should—as the digital methods dictum goes—“follow the medium.” That is, such research should make use of the features of the medium (links, tags, timestamps, and the like) that invite new forms of inquiry. Finally, we have shown that a digital methods approach, although it will not necessarily provide conclusive answers, raises relevant questions that can elicit additional research.ReferencesAng, Ien. Desperately Seeking the Audience. London: Routledge, 1991.Baym, Nancy. “Data Not Seen: The Uses and Shortcomings of Social Media Metrics.” First Monday 18.10 (2013). 23 Sep. 2015 ‹http://firstmonday.org/ojs/index.php/fm/article/view/4873/3752›.Bratich, Jack. “Affective Convergence in Reality Television: A Case Study in Divergence.” Flow TV: Television in the Age of Media Convergence. Ed. M. Kackman, M. Binfield, M. Payne, A. Perlman, and B. Sebok. New York: Routledge, 2011. 55–74.Jenkins, Henry. Convergence Culture. New York: New York UP, 2006.Lee, Hye Jin, and Mark Andrejevic. “Second-Screen Theory: From the Democratic Surround to the Digital Enclosure.” Connected Viewing: Selling, Streaming, & Sharing Media in the Digital Era. Eds. Jennifer Holt and Kevin Sanson. New York: Routledge, 2014. 40–61.Napoli, Philip M. “The Institutionally Effective Audience in Flux: Social Media and the Reassessment of the Audience Commodity.” SSRN Electronic Journal (2013). 23 Sep. 2015 ‹http://papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID2260925_code548166.pdf?abstractid=2260925&mirid=3›.Proulx, Mike, and Stacey Shepatin. Social TV: How Marketers Can Reach and Engage Audiences by Connecting Television to the Web, Social Media, and Mobile. Hoboken, NJ: John Wiley & Sons, 2012. Rogers, Richard. Digital Methods. Cambridge, MA: MIT Press, 2013. SKO. “Kijkcijfers.” Home—Kijkonderzoek. n.d. 23 Sep. 2015 ‹https://kijkonderzoek.nl/kijkcijfers›.Smythe, Dallas W. “Communications: Blind Spot of Western Marxism.” Canadian Journal of Political and Social Theory 1.3 (1977): 1–27.The Nielsen Company. “C3 TV Ratings Show Impact of DVR Ad Viewing.” What People Watch, Listen to and Buy. Oct. 2009. 23 Sep. 2015 ‹http://www.nielsen.com/us/en/insights/news/2009/c3-tv-ratings-show-impact-of-dvr-ad-viewing.html›.———. "Weekly Top Ten." Nielsen Social. n.d. 23 Sep. 2015 ‹http://www.nielsensocial.com/nielsentwittertvratings/weekly/›.Vanderbilt, Tom. "The New Rules of the Hyper-Social, Data-Driven, Actor-Friendly, Super-Seductive Platinum Age of Television." Wired, Mar. 2013. 23 Sep. 2015 ‹http://www.wired.com/2013/03/nielsen-family-is-dead/›.Van Es, Karin. “The Perks and Perils of Social TV: On the Participation Dilemma in NBC’s The Voice.” Television & New Media (forthcoming).