“Yes We Can”: How Online Viewership, Blog Discussion, Campaign Statements, and Mainstream Media Coverage Produced a Viral Video Phenomenon

Online videos have become an important part of the way that members of the public participate in and learn about the political process in the United States.Footnote ¹ Sites such as YouTube, Metacafe, and Daily MotionFootnote ² have become popular placesFootnote ³ for Internet users to not only upload their own politically oriented videos but to view political content posted by others. A recent survey of the online activities of the American public, for example, found that 8 percent of adult users have uploaded a video to a Web site where other people can watch it (CitationMadden, 2007), and YouTube's news and political director has estimated that nearly ten hours of video are uploaded to YouTube every minute (CitationGrove, 2008).Footnote ⁴ Similarly, a recent study by the Pew Internet & American Life Project found that 27 percent of Internet users have gone online to watch speeches or announcements by candidates, 26 percent have watched online videos of interviews with candidates, and 25 percent have watched campaign-related videos that did not come from a news organization or the campaigns themselves (CitationSmith & Rainie, 2008).Footnote ⁵

Online videos have also become an important tool for candidates to use in their efforts to win elections. Beginning with the 2008 presidential primary campaign, a strong presence on YouTube became a prerequisite for any serious candidate running for national political office.Footnote ⁶ Indeed, every major party candidate for president created a YouTube channel to post campaign-related videos, and John Edwards, Hillary Clinton, and Barack Obama even used YouTube videos to announce their candidacies. During the general election, both Barack Obama and John McCain devoted a significant amount of time and energy to communicating through YouTube—with Obama posting over 1,800 videos on the site and McCain posting over 300. Presidential hopefuls, however, were not alone in relying on the site to get their message out. Nearly 72 percent of Senate candidates and over 28 percent of candidates for the House created YouTube channels before the elections in November (CitationWilliams & Gulati, 2009).Footnote ⁷

The online political videos posted by members of the public and candidates for office can gain widespread popularity when they are passed from person to person via e-mail, instant messages, and media-sharing Web sites. If these so-called “viral videos”Footnote ⁸ are frequently discussed in the blogosphere, supported by a candidate's campaign, and widely covered in the mainstream media, they can exert a strong influence on the dynamics of elections. Jim Webb's campaign for the Senate in 2006, for example, received a huge boost in the polls, attracted support from the Democratic National Committee (DNC), and tripled its campaign contributions after posting a video of George Allen calling an Indian-American man a “macaca” on YouTube (CitationScherer, 2006).Footnote ⁹ Similarly, an anonymously producedFootnote ¹⁰ YouTube video (entitled “Vote Different”) in which a woman wearing an Obama campaign logo throws a sledgehammer through a screen playing a clip of Hillary Clinton attracted over 3 million views in less than a month, received widespread attention from national news outlets, and prompted a series of statements by both the Clinton and Obama campaigns (CitationMarinucci, 2007).Footnote ¹¹

Despite their ability to transform election campaigns, there has been almost no systematic empirical research on the factors that lead viral videos to spread across the Internet and permeate into the dominant political discourse.Footnote ¹² As a result, little is known about how online viewership, blog discussion, campaign statements, and mainstream media coverage interact to produce viral political videos. Do journalists, for example, play a critical role in creating a viral video phenomenon by discussing the clips that they see during their own searches of the Internet—regardless of how many other viewers the videos have attracted—or do they only cover an online political video after it achieves some critical mass of viewers? Similarly, do bloggers lead the way in producing viral videos—by channeling online traffic to and alerting journalists about new and interesting videos—or do they merely follow the pack of Internet users and mainstream media reporters who find online videos on their own?

In an initial effort to specify exactly how online political videos “go viral” and change the dynamics of political campaigns, this article asks: What is the relationship between audience size, blog discussion, campaign statements, and mainstream media coverage of an online political video? Using vector autoregression to analyze the data from the most popular viral political video of the 2008 primary campaign—will.i.am's “Yes We Can” music video—I find strong evidence that the relationship between these variables is complex and multidirectional. More specifically, I find that bloggers and members of the Obama campaign played crucial roles in convincing people to watch the video and in attracting media coverage, while journalists had little influence on the levels of online viewership, blog discussion, or campaign statements. Bloggers and campaign members, in other words, seem to occupy a unique and influential position in determining whether an online political video goes viral.

THE MEDIA, THE BLOGOSPHERE, ELECTION CAMPAIGNS AND THE PROCESS OF “GOING VIRAL”

Despite the dearth of empirical research on the specific process of “going viral,” large bodies of literature on the media, the blogosphere, and election campaigns provide a number of clues about what might produce a viral political video.Footnote ¹³ In this section, I draw upon these studies to spell out some initial expectations for the interactions between online viewership, blog discussion, campaign statements, and mainstream media coverage. Based on the existing research, I predict that online views, blog discussion, campaign statements, and mainstream media coverage are likely to be deeply intertwined and each probably exerts a strong influence on the others (see Figure 1). Put differently, a viral political video is most likely the result of a complex and multidirectional interplay between the actions of Internet users, bloggers, campaign members, and journalists.

FIGURE 1. Predicted relationships between online viewership, blog discussion, campaign support, and media coverage.

Display full size

THE “YES WE CAN” VIRAL MUSIC VIDEO

In order to explore whether the relationship between online viewership, blog discussion, campaign statements, and mainstream media coverage is, in fact, complex and multidirectional, I analyzed the “Yes We Can” music video. I chose the “Yes We Can” music video as a case study for exploring the dynamics of viral videos because it was the most popular and high profile online political video of the 2008 campaign. The video, which included cameo appearances by celebrities such as John Legend, Herbie Hancock, Scarlett Johansson, Kareem Abdul-Jabbar, and Kate Walsh, featured black and white clips of Obama's concession speech following the New Hampshire primary set to music written by will.i.am of the hip hop group The Black Eyed Peas. The video was completely “supporter-generated”—with the Obama campaign playing no role in its production.

After debuting the video on ABCNewsNow's “What's the Buzz” on February 1, the video's producers released the video on YouTube,Footnote ²⁴ Dipdive, and a newly created site dedicated exclusively to the video, http://www.yeswecansong.com, early on February 2. Versions of the video quickly spread across YouTube and, within three days, there were over 50 different postings of the video on the site. By the time Obama secured the nomination, the video had been viewed over 20 million times on various Internet sites, inspired a number of widely viewed spoofs,Footnote ²⁵ and been awarded an Emmy for Best New Approaches in Daytime Entertainment (CitationReuters, 2008).Footnote ²⁶

DATA

Systematically studying the dynamic relationship between online viewership, blog discussion, campaign statements, and mainstream media coverage requires daily measures of each variable over a significant period of time.Footnote ²⁷ In order to measure the number of views the “Yes We Can” video received during its first month online (February 2–March 2), I relied on data from Tubemogul.Footnote ²⁸ Tubemogul is a Web site that aggregates video viewing data from sites such as YouTube, Metacafe, and Daily Motion. Tubemogul allows users to track any video they want and provides daily data on the number of times the video was viewed, the number of comments the video received, the number of ratings that were given, and the overall ratings scores. As such, Tubemogul is an excellent resource for scholars interested in studying the rise, spread, and impact of online political videos. Using the unique viewership data provided by Tubemogul's tracking software, I was able to gather data on the number of views the official (and most popular) posting of the “Yes We Can” video received each day on YouTube.Footnote ²⁹

There is no immediately obvious way to measure the amount of blog discussion of an online video, and, as a result, there are important questions about how to proceed in tracking the videos that are given attention by bloggers. Which blogs, for example, should be used to gather data on the videos of interest—only A-list blogs, only less popular blogs, or a mix of both? Similarly, how should “discussion” of a video be measured—by mentions of the title, by mentions of certain keywords, by use of links to the video, or by some other factor?

Following the number of links that lead to the video is likely to provide a better measure of blog discussion of online political videos than keyword searches for three reasons. First, bloggers rarely use the complete title of a video in their post, and, as a result, keyword searches based on the title of an online video can systematically underestimate the level of blog discussion related to it. Second, searches for videos that have common titles, such as Obama's frequently repeated phrase “Yes We Can,” can systematically overestimate the amount of blog discussion about an online video because it will produce many posts that have nothing to do with the video. Finally, following the links that bloggers use is likely to produce accurate estimates of blog discussion because bloggers are likely to link to the sites they are discussing—particularly when the site contains a video.

For these reasons, I decided to use TechnoratiFootnote ³⁰ to track the number of blog links to various versions of the “Yes We Can” video for each day of the study.Footnote ³¹ I decided to use Technorati not only because it collects data on over 110 million blogs but also, and more importantly, because it provides a URL search function that allows researchers to easily track the links that bloggers use in their posts on a day to day basis.

To assess the number of campaign statements about the video, I employed two separate measures. First, I recorded the number of times various versions of the videoFootnote ³² were linked to on the Obama campaign's official blogFootnote ³³ . Second, I tracked the number of times the video's title—“Yes We Can”—and its artist—will.i.am—were both mentioned in e-mails from the Obama campaign.

In order to measure mainstream media coverage of the video, I used a daily count of the number of articles printed in “U.S. newspapers and wire services”Footnote ³⁴ and the number of stories aired on local and national news broadcasts that mentioned both “Yes We Can” and “will.i.am.”Footnote ³⁵ As a result of the fact that print and broadcast coverage were closely related (r =.61), I decided to create an index of overall media coverage for the video.Footnote ³⁶ The Cronbach's alpha for this index was .76.Footnote ³⁷

RESULTS

Some elements of the relationship between online viewership, blog discussion, campaign statements, and media coverage can be discerned by simply looking at how each of these variables changed during the video's first month online. As Figure 2 shows, the video was an instant hit—drawing over 150,000 views on February 2. The audience for the video grew each of the next three days and reached its peak on February 5—when it was viewed nearly 600,000 times. The number of views steadily declined over the next few weeks without any major surges in views. Overall, the official version of the video was viewed over 5.4 million times during its first month on YouTube.

FIGURE 2. Online viewership, blog discussion, and media coverage of “Yes We Can.”

Display full size

Figure 2 suggests that many of the views the video received may have been the result of bloggers linking to it. As Figure 2 shows, well over 1,000 bloggers linked to the site as soon as it appeared on February 2. The number of blog links to the video remained high over the next two days and then started falling off dramatically. As Figure 2 also shows, the number of daily blog posts linking to the video never exceeded 300 after February 6—the same day that video views began declining. Views of the video and blog discussion, in other words, followed very similar patterns of rise and decline.

The Obama campaign also seems to have also played a role in alerting Internet users to the video. Two posts on the campaign's blog linked to the video on February 2, and two days later, on the eve of the Super Tuesday primaries, Michelle Obama sent an e-mail to supporters of the campaign that said, “Sharing this video, which was created by supporters, is one more way to help start a conversation with your friends, family, coworkers, and anyone else who will be voting soon about the issues important to them in this election.”Footnote ³⁸ Unlike bloggers, however, the Obama campaign's attention to the video was decidedly short lived. As Figure 2 shows, the campaign did not mention the video in any blog posts or official emails after February 5.

The online buzz created by bloggers, the support of the Obama campaign, and the increasingly large number of people who had seen the video seems to have quickly captured the mainstream media's attention. After ignoring the video on February 2 and February 3, both the print and broadcast media picked up the story of the video and began discussing its electoral implications. The number of stories rose and remained high until steeply declining on February 8. During the rest of February, the mainstream media periodically ignored and then discussed the video in its coverage of the primary elections. As Figure 2 shows, however, the ebb and flow of media coverage looks very similar to the ebb and flow of viewership and blog discussion on a two day lag.

While looking at the time series graphs presented in Figure 2 provides a general sense of how online viewership, blog discussion, campaign statements, and mainstream media coverage interact, I also used vector autoregression (VAR) to better specify the complex interplay between these variables.Footnote ³⁹ VAR models use lagged values of all of the variables in a system of interrelated variables to predict the current value of each variable in the system (CitationBartels, 1996).Footnote ⁴⁰ This approach is attractive for my purposes here because VAR models, unlike structural equation models, relax a priori assumptions about the direction of causality between variables and the number of time lags to be included in the analysis.Footnote ⁴¹ Indeed, CitationWood and Peake (1998) suggest that VAR is an effective methodology for determining causal relationships when theory is unclear or underdeveloped.

The first step in VAR analysis is to determine the appropriate number of lags to include in the system of equations that is being estimated.Footnote ⁴² Following CitationSims (1980), I determined the number of lags to include in each model by sequentially adding lags to the system of equations and testing for the statistical significance of each additional lag using a modified F-test. Additional lags need to lead to a significant improvement in the fit of the VAR model in order to be included.Footnote ⁴³ Based on Akaike's Information Criterion (AIC) and the Final Prediction Error (FPE), as well as degree of freedom considerations, I selected a lag period of two days.

The next step in VAR analysis is to conduct “Granger causality” tests in order to detect the causal relationships that exist between the variables in the system of equations.Footnote ⁴⁴ “Granger causality” is based on the idea that “variable X causes another variable Y, if by incorporating the past history of X one can improve a prediction of Y over a prediction of Y based solely on the history of Y alone” (CitationFreeman, 1983, p. 338). Granger causality tests, therefore, provide statistical evidence for whether lags of one variable “Granger cause” any of the other variables in the system. More specifically, a chi-squared statistic is used to test the null hypothesis that the lags of the independent variables are significantly different from zero. A significant chi-squared test means that the independent variable “Granger causes” the dependent variable, while an insignificant chi-squared test means that the independent variable does not “Granger cause” the dependent variable.

In order to determine the factors driving the “Yes We Can” viral video phenomenon, I conducted a Granger causality test for the four-equation system that included my measure of online viewership (the number of views for “WeCan08’s” video on YouTube), my measure of blog discussion (the number of bloggers linking to the video), my measure of campaign statements (the number of posts linking to the video on the official Obama campaign blog and the number of e-mails from the Obama campaign mentioning the video), and my measure of media coverage (the index of the number of print and broadcast stories citing the video's title and its maker).Footnote ⁴⁵ The results from the VAR analysis are presented in the Appendix. Table 1 displays the results of the Granger causality test. Each dependent variable is listed in the first column along with all of its independent variables in the second column.Footnote ⁴⁶ Figure 3 presents the same information in a slightly different form—a diagram showing the observed relationships between each of the variables.

FIGURE 3. Observed relationships between online viewership, blog discussion, campaign support, and media coverage.

Display full size

The Granger causality test results presented in Table 1 and in Figure 3 reveal a number of interesting relationships. First, support from the Obama campaign was crucial in making the “Yes We Can” video go viral. As Table 1 and Figure 3 show, the number of statements made by the campaign exerted a significant influence on the size of the online audience, the amount of discussion in the blogosphere, and the number of media stories about the video. When the campaign sent e-mails and posted messages on their official blog about the video, Internet users, bloggers, and journalists seem to have taken this as a cue that the video was something worth paying attention to. Although, as suggested above, there were many reasons to expect that campaigns could produce these kinds of effects on viewership and discussion of online political videos, these findings do provide the first empirical proof that campaigns can function as more than passive bystanders when their supporters create videos and post them online. Simply put, these results suggest that campaigns can make all of the difference in transforming supporter-generated videos from undiscovered white noise to a viral video phenomenon.

Second, bloggers also played an important role in drawing the attention of Internet users and journalists to “Yes We Can.” As Table 1 and Figure 3 show, blog discussion exerted a significant impact on both the number of people who watched the video and the amount of media coverage the video received. The fact that blog discussion produced more views and more news coverage of the video should not be surprising. Indeed, as suggested above, a large body of research shows that bloggers can have a major impact on what media chooses to cover (CitationBloom, 2003; CitationDrezner & Farrell, 2004; CitationRoth, 2004; CitationSmolkin, 2004; CitationWallsten, 2007), and the links provided in blog posts can drive up viewership statistics by making it easy for interested readers to locate and watch the video online. The fact that blog discussion of “Yes We Can” significantly influenced online viewership and media coverage is yet another example of the increasingly important role that bloggers are playing in structuring media coverage of and public opinion about political events. These findings do, however, make a strong case that online political video makers should cater to the interests of bloggers if they want their videos to attract a large audience in both the online and offline political worlds.

Third, the large number of views the “Yes We Can” video attracted with Internet audiences was an important part of its ability to attract attention beyond the online world. Indeed, the Granger causality test results presented in Table 1 and Figure 3 show strong evidence that the size of the online audience influenced the amount of mainstream media coverage “Yes We Can” received and the extent to which the Obama campaign supported the video. Interestingly, the number of views the video received did not significantly affect the level of discussion about the video in the blogosphere. Based on this evidence, it appears that journalists and members of political campaigns may be carefully tracking the popularity of the content posted on sites such as YouTube, Metacafe, and Daily Motion in order to determine what videos they should discuss, while bloggers tend to base their decisions about what to talk about on other considerations, such as what campaigns are saying. Put differently, although bloggers may discuss online political videos in the absence of gaudy viewership statistics, a large audience seems to be a necessary condition for journalists and campaign members to devote time and energy to discussing an online political video.

Finally, although journalists covered “Yes We Can” extensively during its first month online, there is no evidence that media reports contributed to the video going viral. As Table 1 and Figure 3 show, media coverage failed to exert a significant influence on blog discussion, campaign statements, or online viewership. This finding was highly unexpected; all of the previous literature on the media's agenda-setting power suggested that widespread news coverage of an event such as the video would lead members of the public to seek it out online, bloggers to blog about it on their blogs, and the campaign to discuss it in their official statements. Of course, the fact that media coverage was not a significant influence on blog discussion, campaign statements, or online viewership does not mean that the media is entirely unimportant when it comes going viral. There is no doubt, for example, that television broadcasts and newspaper stories about “Yes We Can” increased the overall number of people who learned about and saw the video. These findings do, however, establish that journalists are likely to be followers rather than leaders in the process of creating viral political videos.

CONCLUSION

This article began by asking: What is the relationship between audience size, blog discussion, campaign statements, and mainstream media coverage of online political videos? Using VAR to analyze the data from the most popular viral political video of the 2008 campaign—will.i.am's “Yes We Can” music video—I found strong evidence that the relationship between these variables is complex and multidirectional. More specifically, I found that bloggers and members of the Obama campaign played crucial roles in convincing people to watch the video and in attracting media coverage while journalists had little influence on the levels of blog discussion, online viewership, or campaign statements. Bloggers and campaigns, in other words, seem to occupy a unique and influential position in determining the whether an online political video goes viral.

In addition to providing the first glimpse into the dynamics that create a viral political video, the findings presented in this article make contributions to three separate bodies of research. First, the conclusions presented here contribute to the emerging literature on the consequences of political blogging by suggesting that the interest of bloggers is a central factor in explaining the rise, spread, and decline of viral videos. Second, the conclusions presented here contribute to the literature on media agenda-building by detailing the factors that drive journalists to cover developments in the online world. Finally, the conclusions presented here contribute to studies of political campaigns by highlighting the role that campaign actions can play in promoting supporter-generated content online.

Future research should build on the findings presented here in a number of ways. First, studies of other political videos are needed in order to determine whether online views, blog discussion, campaign statements, and mainstream media coverage interact with and influence each other in the same ways that I have found here. There are many reasons to expect that the dynamics driving the “Yes We Can” video were unique—it was filled with celebrity entertainers, it was first played on a mainstream media organization's broadcast, and it became popular almost immediately after it was posted on YouTube. For online political videos that do not have this distinct set of characteristics, audience size, blog discussion, campaign statements, and media coverage may influence each other in different ways. The number of online views, for example, may be an important factor in driving blog discussion when no celebrities are featured in the video, because it can provide a compelling reason for talking about the video. Similarly, blog discussion may not exert such a strong influence on media coverage of online political videos when the number of people who have viewed the video is not rising as quickly as it did in the case of the “Yes We Can” video, because blog discussion alone may not be enough to warrant coverage. In more formal terms, the case study presented here may lack external validity. Future studies, therefore, need to sample a larger number of online political videos in order to get a better, more general sense of how audience size, blog discussion, campaign statements, and mainstream media coverage influence each other.Footnote ⁴⁷

Second, in order to gain a fuller understanding of the ways that viral political videos emerge, spread, and influence elections, data on how online political videos are passed from person to person through e-mail, instant messages, and social networking sites such as MySpace and Facebook is needed. While blog discussion, media coverage, and campaign statements are certainly important in diffusing online political videos, most people who find out about an online political video probably do so through e-mail, instant message, and social network communication with their friends and family. Recent surveys by the Pew Internet & American Life Project, for example, have found that 75 percent of online video viewers have received links to online videos via e-mail or instant messages (CitationMadden, 2007) and that 9 percent of Internet users have forwarded or posted someone else's political recordings using e-mail or instant messages (CitationSmith & Rainie, 2008). Similarly, CitationRobertson, Vatrapu, and Medina (2009) found that links to YouTube were more common than links to any other site on the Facebook “walls” of presidential candidates during the 2008 campaign—suggesting that social networking sites may be an important mechanism by which Internet users inform each other of new and exciting online political videos. Because I did not include measures of these potentially important factors in this analysis, the statistical findings presented here could be biased and misrepresent the true relationship between online viewership, blog discussion, campaign statements, and media coverage. Future work on the relationship between these variables should, therefore, try to include measures of e-mail, instant message, and social network diffusion in order to ensure accurate estimates about the complex influences that online viewership, blog discussion, campaign statements, and media coverage exert on each other.Footnote ⁴⁸

Third, given the central role that bloggers play in the creation of viral political videos, future research should explore the factors that lead bloggers to discuss certain online videos and to ignore others. One potentially interesting line of inquiry would be to explore the role that ideology and partisanship play in the linkage patterns of bloggers. A number of recent studies have suggested that political bloggers link primarily to sites that share their ideological predispositions. CitationAdamic and Glance (2005), for example, analyzed linkage patterns among popular political blogs during the final two months of the 2004 election campaign and found a great deal of fragmentation along ideological lines—with conservative bloggers being much more likely to link to other conservatives and liberal bloggers being much more likely to link to other liberals. Similarly, CitationHargittai, Gallo, and Kane (2008) examined links between A-list political blogs during three one-week periods in 2004 and found that conservative and liberal bloggers are vastly more likely to link to blogs that share their point of view. Future research should explore the extent to which bloggers link to videos that confirm their predispositions and, more importantly, whether ideological or partisan linking tendencies have any influence on the role that bloggers play in spreading online political videos.

Finally, future research is needed to assess the impact that viral political videos have on members of the mass public. As suggested above, viral videos are a potentially important influence on electoral campaigns not only because so many people watch them but also because bloggers and journalists discuss them in their coverage of political events. In addition to being viewed nearly six million times, for example, the “Yes We Can” video was linked to on over 7,000 blog posts and discussed in nearly 400 print and broadcast news stories in the first month after its release. Unfortunately, little is known about how all of this attention influenced the attitudes and behaviors of members of the public. Did exposure to media coverage of “Yes We Can” lead undecided voters to watch the video and support Barack Obama? Did blog discussion of the video motivate supporters to contribute money or volunteer their time to the Obama campaign? Ultimately, the true importance of viral videos cannot be fully assessed until systematic empirical research on individual level questions such as these is completed.