You are on page 1of 37

Social Media in the 2012 Presidential Election: Revolution or Reflection?

Abstract: The role of the Internet in elections and political discourse has been hotly debated. This paper looks at tweets, photos and videos sent out by the Barack Obama and Mitt Romney campaigns over the last 90 days of the 2012 United States presidential election. Through a content analysis of both manifest and latent content, the paper demonstrates that substantial differences exist in the specific policy issues the campaigns talked about the most, but their organisational efforts were relatively similar. Moreover, the paper finds that the style and substance of the campaigns messages varied over the different platforms. Key words: political discourse, social networks, e-democracy, Facebook, Twitter, YouTube, content analysis

Introduction Different forms of media have long had an impact on the outcome of elections: John F Kennedys visually commanding appearance in the 1960 US presidential debate and the Sun newspapers claim to have won the 1992 British general election for the Conservative Party are two famed examples of the influence that traditional media hold. More recently, the mass adoption of the Internet has presaged a new debate about the role that online forms of communication might play in political campaigns.

The US presidential election of 2012 saw an increasingly sophisticated use of online communication. The soaring popularity of online social networks like Facebook, Twitter and YouTube amongst ordinary Americans led both the Obama and Romney campaigns to establish a substantial presence on these sites, using their inherent interactivity as a means to promote the candidates political platform and mobilise fundraising, volunteering and get out the vote efforts.

Of course, the 2012 campaign represents only one empirical example of the everchanging interplay between the Internet and political systems. But judging by the statistics, it is a prominent one. Barack Obamas Facebook page (http://www.facebook.com/barackobama) has over 35 million likes, a tweet heralding his election victory became the most retweeted of all time (Stableford, 2012), and both campaigns had digital teams with more than 100 staff (Inside the Cave, 2012, p.11). The sheer size and scale of the campaigns social media efforts, therefore, suggests that researching the 2012 election may provide

interesting lessons about the use of social media in political campaigns more broadly.

Background and rationale

In investigating the role of social media in the 2012 presidential election, this study builds above all upon two strands of research. The first pertains specifically to how party machines have used the Internet to their electoral advantage. In a review article, Parties in the digital age, Gibson et al (2009) describe and distinguish between two ways in which political parties might draw upon digital technology. The first grassroots strategy, they suggest, sees ICTs forming a new conduit for mobilising membership, whilst the second approach describes Internet technologies harnessed toward marketing an increasingly hollow memberless party to voters (p.89).

This is an important distinction. Many prominent pieces of research into political parties and digital technologies, such as Rasmus Kleis Nielsens ethnographic study of two congressional campaigns in Ground Wars (2012), have focussed on the improved grassroots mobilisation that the development of more sophisticated information systems have allowed. Yet whilst most empirical work has focussed on the use of such grassroots strategies, the idea of technology helping to create the hollow memberless party model was the subject of Helen Margetts paper on The Cyber Party (2006). Margetts postulated that future parties may develop looser ties with voters, abandoning the strict membership

structure that characterised the growth of institutionalised parties in the twentieth century.

This research draws upon Margetts theory to investigate the increasingly loose relationship between campaigns and voters, neatly characterised by the uptake of online social networks amongst politicians and parties. Debate centres on whether, as Margetts claims, democracies built on the painless click of a button may be more vibrant than traditional systems (p.21) or whether they will ultimately be less effective and engaging, as argued by Malcolm Gladwell (2010).

To seek to answer this question, a better understanding is needed of precisely how campaigns use popular social networks to interact with prospective voters, something this study seeks to provide. This links to the second strand of research upon which this study builds, namely how political discourse is conducted between both ordinary people and political elites on emerging social networks.

Foremost amongst research in this realm is Twitter. The microblogging site has become especially popular amongst journalists other users as a platform on which to spread news, share links and have shorthand political discussion. However, the exact nature of the site, and what its popularity means for political discourse, has been keenly debated. In particular, a key question is whether Twitter represents the logical conclusion of the echo chamber phenomenon, described by writers such as Cass Sunstein (2007) and Eli Pariser (2011), whereby users filter out political views which do not correspond to their own,

thus reinforcing pre-existing beliefs and ultimately making political discourse more extreme and bipolar. The empirical evidence for this is mixed: a study by Conover et al (2011) found ideological clustering prevalent in the retweet functionality but significantly less so in replies to tweets. Studies have also looked at the role played by Facebook in political engagement: Bond et al (2012) found that information about whether friends have voted can have a small effect on a users own voting behaviour. Finally, looking more broadly, Gentzkow and Shapiro (2011) have found that the Internet features relatively less ideological segregation than offline platforms.

This study seeks to fill in some of the gaps in existing research by assessing the use that high-level political campaigns have made of various social networks. By investigating, for example, which policy issues were covered, what sort of organisational appeals to voters were made, and how output varies across different platforms, the study will use presidential campaigns as a springboard for a wider discussion and future study about different social networks and their political impact.

Research question

The two main strands of research that this study builds upon are focussed around two overarching questions. Firstly, will increasing democratic engagement on social networks lead to a more vibrant or a more apathetic democracy? And secondly, do the technical architecture and usage norms on popular social networks like Twitter lead to greater ideological segregation?

Answering these major questions will likely require multitudinous study over a long period of time; this study is more modest. The specific research question this study seeks to answer is:

How did the Obama and Romney campaigns use Twitter, Facebook and YouTube to promote their candidate and mobilise supporters?

Through answering this question, it is hoped that wider lessons can be drawn about the uses of different social networks for different purposes during elections, and that tentative judgments can be drawn about how differences in technical architecture and usage norms affect campaign activity.

Methodology

Approach

This study adopts a primarily qualitative approach to the study of the campaigns use of social media during the 2012 election. As Robson has noted, process questions [which] focus on how things happen [tend] towards a qualitative approach (2011, p.61), and the research question posed above would seem to fit this model of a process model. However, inherent in the research question is also the suggestion of differences in the campaigns use of the various social networks, which posits a relationship to be investigated. This sort of inquiry, or what Robson calls variance questions, tend toward the use of a quantitative

approach (p.61). This study thus involves both exploratory and explanatory elements and therefore uses content analysis, an analytical method that encompasses both qualitative and quantitative components.

Data sources, assumptions and scope

Most studies of the role played by social networks in political discourse, including those cited above, have tended towards quantitative methodologies, typically analysing a large corpus of data. (The title of the Bond et al (2012) study, A 61-million person experiment in social influence and political mobilization is a prime example.) This study takes a different approach, studying just two accounts, those of Barack Obama and Mitt Romneys campaigns, on three social networks, Twitter, Facebook and YouTube. The scope of the study across time also needs to be defined. As Bryman has noted, sometimes the decision about dates is more or less dictated by the occurrence of a phenomenon (2012, p.293), and that is the case in this instance, as the study centres on the communication during the presidential election. Nonetheless, the US election process has lengthened considerably over time, and has no official starting date in the manner of, for example, UK general elections, which officially begin when Parliament is dissolved. As such, a somewhat arbitrary marker had to be introduced, and the date selected was 9th August 2012, the day before Paul Ryan was named as Mitt Romneys running mate on the Republican ticket. This meant that 90 days worth of data was collected from each source.

This research design comes with its own strengths and drawbacks in relation to larger-scale, quantitative studies. A core strength is that the dataset is just small enough that meaningful, qualitatively-derived codes could be generated for coding (see Analytical approaches below).

However, this approach necessitated a number of assumptions about the data that do not apply to larger-scale studies. Firstly, the data sources were obviously selected on an entirely unrepresentative, a priori basis. A study of the Obama and Romney Twitter accounts, for example, would not be expected to offer much of an explanatory basis for the average users Twitter experience, whether in terms of the traditional idea of external validity, or in Lincoln and Gubas (1985) notion of transferability as a criterion for qualitative research. It is important to note, however, that even research which compiles data on a statistically representative basis from a social network such as Twitter cannot abstract findings to broader populations, as the Twitter population is a highly nonuniform sample of the population (Mislove et al, 2011, p.1). External validity can therefore prove problematic for both qualitative and quantitative studies, depending on the scope of findings.

Another core assumption that this study must make relates to the authors of the data being analysed. It must be assumed that the Obama and Romney campaigns used social networks in a highly sophisticated and reflexive manner, and that their ultimate aim in doing so was winning votes for their candidates. This represents a clear contrast to ordinary uses of social networks, which might involve less self-awareness and lack the clear, singular focus of winning an

election. Again, however, the distinction raised here should not be exaggerated. Viewed through the lens of Uses and Gratifications Theory, which states that users of a particular technology use it for a specific need, it might be seen that every user of a social network has particular needs be they social, professional or otherwise which they seek to meet through the network. Moreover, this is by no means the first piece of research to investigate users of social media with nakedly political motives: a number of studies have looked at the behaviour of political bloggers, for example (Drezner and Farrell 2008, McKenna and Pole 2008, Hargattai et al 2008). Therefore, whilst the assumption of sophistication and reflexivity in the campaigns social media usage must be made, this does not preclude drawing valid findings within the scope of the study as outlined.

Procuring data and establishing quality

Whilst choosing the actual sources of data to be used was a straightforward process, there were a number of issues associated with specifically what data should be procured. The first question concerned exactly what data should be recorded. In the case of Twitter, this was relatively simple: tweets are by definition short, text-based posts. As such, the Twitter data that was downloaded was simply constituted by the author (the @BarackObama and @MittRomney accounts), the content of the tweet and the date and time it was sent. The unit of analysis was therefore the tweets themselves. There was however another issue to be considered in the case of storing tweets. A majority of the tweets had, as part of the 140 character limit, short links to external sites containing more information, usually on the candidates websites (see Results and discussion).

This presented a methodological dilemma: should the content of the links be considered alongside the tweets themselves? In light of the typical user experience on Twitter, it was decided that only the fact that a tweet had a link, not the content of that link, would be coded. It did not seem accurate or appropriate to code content existing outside the source of data itself.

In the case of Facebook, it was decided that the data to be analysed would be photos. Visual images constitute a large proportion of Facebook content, and this was especially so in the case of the Barack Obama and Mitt Romney Facebook pages. Almost every single Facebook post from both accounts during the 90-day period featured some form of visual imagery accompanying the text. Focussing on the photos each campaign posted on Facebook therefore provides an interesting contrast to the text-based Twitter updates. However, a wide range of different imagery was employed by both campaigns, including infographics and text overlays as well as pure photos. It was therefore decided that for the sake of consistency in the analysis, only photos without any text overlay would be recorded for analysis. However, any text that accompanied the photos in the caption would also be coded for, as it was reasonable to expect that most viewers who saw the photos would also see and read the text. As with the issue surrounding Twitter links, the approach used here was considering what the typical viewer might be expected to have seen.

Selecting the data to be used on YouTube was similar to that for Facebook. YouTube is centred on allowing users to upload and watch videos. It therefore stands to reason that the primary content the Obama and Romney campaigns

10

would seek to propagate on YouTube would be videos. As with photos on Facebook, however, the description beneath each YouTube was also coded in conjunction with the video, as again, the description likely forms part of what a typical user might see.

Having identified precisely which data would be recorded from each source, the next concern was storing the data for the purposes of analysis. Each data source came with its own challenges in this regard. The most difficult platform on which to actually access the data was undoubtedly Twitter. Tweets are typically stored for only a matter of weeks, and this research took place after most of the tweets during the time period under investigation had disappeared. Therefore, the third party website Snapbird, which stores tweets for a considerably longer time period than Twitter makes publicly available, was used to download the tweets. Unfortunately, the Snapbird service introduces a degree of unrepresentativeness into the sample itself. Due to the throttling technique employed by Twitters API, not all tweets sent by the campaigns accounts were captured by Snapbird. This introduces a regrettable bias into the data, although it is important to note that this paper makes no findings with a basis statistical significance. Thus while the bias introduced by the way in which tweets were procured is certainly unhelpful, it also serves as a reminder at the outset as to the known empirical limitations of this study. In the case of both Facebook and YouTube, the issue was not accessing the data but storing it, given the large size of the photo and video files. The content was not therefore downloaded as such, but records were made of each photo and video at the same time as the tweets were downloaded

11

to ensure accuracy, and the photos and videos were coded externally from the original hosts, Facebook and YouTube respectively.

The final step in preparing the data for analysis was to create a smaller sample from it. Whilst the amount of data collected was not enormous, given that only two accounts one each site were studied, the data set collected was still large enough that content analysing all of it, particularly the long videos, would have taken up a prohibitive amount of time. Moreover, the amount of data collected for each candidate varied. As Table 1 shows, the Obama campaign posted more tweets and videos, but the Romney campaign had more Facebook photos. In order to make fair comparisons between the different campaigns, especially in discussing the relative frequency that particular codes were used, it was necessary to level out these totals. This was conducted using systematic sampling, which Robson describes as taking every nth name from the population list (2011, p.272). Whilst Bryman notes that it is important to ensure that there is no inherent ordering of the sample frame (2012, p.192) when using systematic sampling, the fact that this sample is ordered by date is not itself problematic. Table 1 shows the original number of tweets, photos and videos collected from each candidates account, and the number of samples drawn from each sample source.

12

Table 1 Obama original Tweets Facebook photos YouTube videos 521 10 133 10 1574 87 240 20 Sampled Romney original 312 161 240 20 Sampled

Analytical approach: content analysis

There are a number of reasons why content analysis is an especially suitable way of approaching the data collected for this study. Content analysis is typically performed on various forms of human communications; this may include various permutations of written documents, photographs [or] motion pictures (Berg 2004, p.304). This flexibility suits perfectly the multimedia dataset under investigation. In addition, as Bryman notes, content analysis can allow a certain amount of longitudinal analysis (2012, p.305), tracking changes in messaging over time, well suited to the ebb and flow of various issues through an election campaign. Additionally, analysing content from different sources allows some comparisons to be drawn between them, comparisons which can be expressed in a quantitative sense.

13

Before conducting the content analysis, however, a number of procedures needed to be defined. Most importantly, precisely what should be counted or coded for needed to be decided upon; this relates directly to research question above, which lists two ways in which social media may have been used by the campaigns: to promote the candidate and to mobilise supporters. This suggested two broad areas for the coding to be framed around: the Political and the Organisational. In the typology developed (See Appendix), the Political category was developed to allow the coding of the substantive political issues in the campaign, for example the economy and the military. As Berg mentions, researchers [can] use some categorical scheme suggested by a theoretical perspective (2004, p.311) as a starting point for developing a coding frame. In this instance, the majority of the codes in the Political category were derived from the Twitter Political Index, a site which sought to discern the key political issues at play in the 2012 election.

The Organisational dimension was designed to answer the second aspect of the question, how the campaigns tried to mobilise supporters. This drew upon the two basic commodities that campaigns seek from supporters, money and time, in the fundraising and timeraising categories, as well as direct appeals for supporters to go and vote, either in person on election day or ahead of time. Finally, the category of online interaction was added, to record an attempt by a campaign to encourage the promotion of a candidate more widely through the use of online networks: for example, encouraging a retweet on Twitter or a share on Facebook.

14

In identifying these political issues and organisational tactics in the data, it became necessary not simply to code for specific words but to look for underlying themes. For example, politicians generally avoid the use of this term abortion, preferring different terminology, so an attempt to merely code the word abortion would not produce an accurate representation of references to the topic. As Bryman notes, then, when the process of coding is thematic, a more interpretative approach needs to be taken (2012, p.297). This relates to the distinction between what is termed manifest and latent content analysis. Coding for latent content not literally visible in the text, then, is likely to involve interpretation on behalf of the researcher interpretation that may also introduce bias or subjectivity. Given that it is its reputation for transparency and replicability that often causes content analysis to be referred to as an objective method of analysis (Bryman 2012, p.304), this presents a potentially dangerous challenge to the rigor of the methodology.

However, there are ways to mitigate the effects of this problem. Berg suggests that researchers should offer detailed excerpts from relevant statements that document the researchers interpretations (2004, p.308); this is done in the Appendix. Moreover, Berg also recommends that perhaps the best resolution of this dilemma about whether to use manifest or latent content is to use both whenever possible (p.309), something undertaken in this study. As such, there are also categories on the coding scheme predicated on manifest content: the Time dimension required the coding of each piece of content based on the date it was published. There was also a category for coding whether tweets had links incorporated into them; again, this is an example of manifest content. The final

15

dimension, whilst relying on latent content such as the inclusion of celebrities and mention of the candidates opponents and families, was relatively straightforward to apply.

Much of the discussion of coding in content analysis relates specifically to textbased content. However, two of the sources of data in this paper are non-textual photos from Facebook and YouTube videos. A chapter on the content analysis of images in Gillian Roses influential book Visual Methodologies suggests that as the process [involves] reducing the rich material in any photograph to a series of codes photos (and by extension videos) should be coded in a way which has some analytical significance (2007, p.60). As such, the coding categories developed specifically for visual imagery in the photos and videos were concerned with the presentation of each candidate. These included coding for whether the photo showed an urban or rural setting, whether large crowds or small groups of supporters were shown, and whether the photos used traditional American iconography such as flags and banners, or showed a more modern multiculturalist flavour. Finally, a code was constructed to investigate whether the longstanding political tradition of candidates being photographed with babies persists.

Discussion of the construction of the coding frame thus far has related to the predefinition of coding categories, which suggests the deductive investigation of whether (and to what extent) certain political issues and organisational motives, generated abstractly, actually appear in the content being analysed. However, this is only one component of the dualistic process of category development. An

16

inductive approach involves researchers immersing themselves in the content in order to identify the dimensions or themes that seem meaningful (Berg, 2004, p.311). Whilst in this study the majority of categories were derived a priori, some emerged inductively out of the coding process itself. Of particular note are the categories in the political dimension which were added, including debt and deficit in response to the Romney campaigns repeated references to the countrys borrowing as a salient issue, and women in response to the Obama campaigns frequent mentioning of womens issues above and beyond abortion.

Limitations

The content analysis approach used in this study encountered various limitations. Firstly, in terms of researcher bias, having had some on-the-ground experience of American politics, including working with one of the campaigns being studied, may have exacerbated the problems with objectivity already highlighted above. More generally, an issue which is perhaps underestimated in the literature is a researchers expertise when coding latent content. Especially on Twitter, given the shortness of posts, particular content may be understood by the researcher in ways that differ from the average audience member. It is perhaps futile to try to identify an average user, but the expertise researchers bring to latent content analysis may have an undue influence on how coding is conducted.

17

Another limitation of this approach was its failure to produce exhaustive and mutually exclusive categories, as recommended by Rose, who does at least acknowledge this is extremely difficult (2007, p.60). Unfortunately, even at the level of a single tweet, multiple political issues were often coded, such that the codes were not mutually exclusive. Some dimensions of categories, however, were more successful in this regard, such as those for coding the visual imagery in photos, including the mutually exclusive rural and urban settings.

Content analysis is also limited as an analytical approach by the scope of its findings. It lacks explanatory power: as Bryman notes, it is difficult to ascertain the answers to why questions through content analysis (2012, p.307). Secondly, although content analysis allows the presentation of findings in numeric terms, Berg warns that researchers must be cautious not to take or claim magnitudes as findings in themselves (2004, p.309). This is especially significant in this study, given that the raw number of tweets, photos and videos had to be levelled for the sake of analysis. With these caveats in mind, this paper now turns towards the major findings of the study.

Results and discussion

Having devised a primarily exploratory piece of research and coded the three types of content using the coding scheme (see Appendix) a number of important findings emerged. The most significant are detailed below.

18

Issue salience: its the economy, stupid

As might have been expected, given the political climate, by far the most frequent substantive issue discussed on the campaigns social media channels appears to be the economy and jobs. This appears to have been central to Romneys message in particular, with over half of his issue-oriented tweets relating to the economy. However, it appears that, even if the economy was the central battleground upon which the election was fought, wedge issues are still prevalent in American political discourse online. A particularly revealing approach here is to compare the political issues each campaign tweeted about with the policy preferences of the American public expressed in a 2012 Pew Research survey (Public Priorities, 2012). The survey found clear partisan divides on a number of policy priorities: for example, the Democrats over Republicans favoured improving [the] educational system reducing health care costs by wide margins, whilst Republicans preferred strengthening the military and reducing the budget deficit. To a surprising degree, the subjects that the Democrat and Republican presidential campaigns in the sampled tweets reflected these partisan divides (see Figure A). The Obama account tweeted about education and health more than ten times each; Romney tweeted about the military 8 times compared with 3 references from Obama, and 16 times about the budget deficit a subject the Obama campaign didnt touch upon once in the sample.

19

Figure A: Code application report from Dedoose for the Political dimension

This finding suggests that presidential campaigns rely on firing up their bases by relying on wedge issues that are likely to inflame the passions of supporters. It is especially interesting that this appears to be occurring on Twitter, seemingly an open, interactive communication landscape compared with, for example, blog networks. Campaigns might therefore view Twitter through the prism of Cass Sunsteins (2007) echo chamber notion, seeing it as an environment for reinforcing existing predispositions as much as changing the minds of moderate voters.

20

Facebook and YouTube: style over substance

In contrast to the highly issue-charged discourse on Twitter, YouTube videos and Facebook photos emerge from the analysis as relatively soft-focus platforms, with more emphasis paid towards biographical details about the candidates. For example, in the run-up to the Democratic National Convention, the Obama Facebook page posted a cache of photos depicting Barack Obamas early life, whilst on YouTube, the Romney campaign published videos promoting their candidates job creation record at Bain Capital. Attacks on an opponents character and policies were also made in YouTube videos, with four of ten videos analysed for both campaigns involving direct references to the opposing candidate. Perhaps in the friendship-oriented world of Facebook, however, ad hominen attack ads may struggle to gain acceptance. This brings home the point that political campaigns must deal with social networks in the way they find them. Each network is already invested with usage norms constituted by millions of ordinary users, and a single political campaign, no matter how big, will likely be successful only by conforming to these norms.

Organisationally, a level playing field

The two sides put a similar emphasis on fundraising, get-out-the-vote efforts and finding volunteers in their social media campaigns. The findings show a similar level of tweets and videos dedicated to these activities (see Figure B). Both sides, too, saw the advantage in utilising the pre-existing networks of followers, by repeatedly encouraging users to retweet this or share this with your friends.

21

Clearly, the technical architecture of social networks, as well as their usage norms, is being incorporated into campaigns appeals to supporters.

Figure B: Code application report from Dedoose for the Organisational dimension

Reflections and recommendations

I believe that these findings are a good demonstration of what can be achieved through a relatively small-scale qualitative study with the use content analysis. It was very satisfying to be able to provide empirical weight to long-held

22

assumptions, such as the findings regarding issue salience. On a more practical level, this study also impressed on me the sheer volume of time required to conduct research of almost any kind or size. The coding itself took a great deal of time, as did procuring and arranging the data, and there were a number of timeconsuming missteps which could have been avoided with better planning. I found the Dedoose software a useful companion, particularly in its presentation of data, but the prohibitive costs associated with uploading photos and videos recommended against using this software for non-textual analysis.

Academically speaking, future studies could dig much deeper into what the campaigns social media approach means for democratic engagement online. Whilst using content analysis was a fruitful approach for a study of this size, a more sophisticated mixed methods study might use interviews with members of the social media teams and a larger-scale, statistically significant data analysis to produce richer and more externally valid findings.

Conclusion

This study sought to investigate the new and often innovative uses that political parties are making of popular social media sites, in the context of the 20 US presidential election. Data was collected from the official campaign accounts of Facebook, Twitter and YouTube. Content analysing this data generated a number of interesting findings: that substantial differences exist in the specific policy issues the campaigns talked about the most, but their organisational efforts were relatively similar. The findings also showed that different social networks were

23

utilised in different ways: Twitter saw the greatest level of policy discussion whilst YouTube and especially Facebook were more focussed on presenting candidates rather hagiographically. It was speculated that these differences in uses may relate to the varying technical architecture and differing user norms in each environment.

This study therefore offers directions for both future campaigns and academic research. In spite of the various methodological limitations described, this research constitutes a positive contribution to the corpus of literature on digitalera politics.

24

References Berg, Bruce L (2004). An Introduction to Content Analysis. In Qualitative Research Methods. London: Pearson. Bond, Robert M., Fariss, Christoper J., Jones, Jason J., Kramer, Adam D. I., Marlow, Cameron, Settle, James E. and Fowler, James H. (2012). A 61-million-person experiment in social influence and political mobilization. In Nature, 489, 295298. Bryman, Alan (2012). Social Research Methods. Oxford: Oxford University Press. Conover, M., Ratkiewicz, J., Francisco, M., Goncalves, B., Flammini, A., & Menczer, F. (2011). Political Polarization on Twitter. Fifth International AAAI Conference on Weblogs and Social Media. Barcelona, Spain. Farrell, Henry and Drezner, Daniel W. (2008). The power and politics of blogs. Public Choice 134 (1): pp.15-30. Gentzkow, M. and Shapiro, J.M. (2011). Ideological Segregation Online and Offline. The Quarterly Journal of Economics 126: pp. 1799-1839. Gibson, Rachel and Ward, Stephen J. (2009). Parties in the Digital Age: A Review Article. Representation, 45(1): 87100. Gladwell, Malcolm (2010). Small Change. Retrieved from http://www.newyorker.com/reporting/2010/10/04/101004fa_fact_gladwell Hargattai, Eszter, Gallo, Jason and Kane, Matthew (2008). Cross-ideological discussions among conservative and liberal bloggers. Public Choice 134 (1): pp. 67-86. Inside the Cave: An In Depth Look at the Digital, Technology, and Analytics Operation of Obama for America. (2012). Retrieved from http://engagedc.com/download/Inside%20the%20Cave.pdf. Lincoln, Y. S. and Guba, E. (1985). Naturalistic Inquiry. Beverley Hills: Sage. Public Priorities: Deficit Rising, Terrorism Slipping (2012). Retrieved from http://www.people-press.org/2012/01/23/section-1-the-publics-policypriorities/. Margetts, Helen (2006). Cyber Parties. In Z. Katz, R. S. and Crotty, W. (Eds), Handbook of Party Politics. London: Sage. McKenna, Laura and Pole, Antoinette (2008). What do bloggers do: an average day on an average political blog. Public Choice 134 (1): pp.97-108. 25

Mislove, A., Lehmann, S., Ahn, Y.-Y., Onnela, J.-P., & Rosenquist, J. N. (2011). Understanding the Demographics of Twitter Users. Fifth International AAAI Conference on Weblogs and Social Media. Barcelona, Spain. Nielsen, Rasmus Kleis (2012). Ground Wars. Princeton: Princeton University Press. Pariser, Eli (2011). The Filter Bubble. United States: Penguin. Robson, Colin (2011). Real World Research. Oxford: Blackwell. Rose, Gillian (2007). Visual Methodologies: An Introduction to the Interpretation of Visual Materials. London: Sage. Stableford, Dylan (2012). President Obama election-night tweet retweeted more than Justin Biebers tweet. Retrieved from http://news.yahoo.com/blogs/ticket/obama-retweet-bieber-record081008623--election.html. Sunstein, Cass R. (2007). Republic.com 2.0. Princeton: Princeton University Press.

26

Appendices Appendix A: Coding Scheme


Title Organisational Online interaction Timeraising Fundraising In person voting Early voting Political Abortion Women Economy Education Gay rights Foreign Policy Healthcare Immigration Retirement Taxes Terrorism Energy and Environment Military Patriotism Deficit and debt Time Convention era Election era Debate era Personal Candidates Opponents Ordinary people Celebrities Miscellaneous Has link In Spanish Is Retweet American Description if applicable Encourages the user to undertake some form of purely online action, however small Encourages viewer to volunteer their time for the campaign Encourages viewer to volunteer their money for the campaign Encourages viewer to register to vote, physically go and vote, or tell others about voting Informs or encourages viewer to vote before Election Day

At what point in the campaign the media was published to September 15, 2012 October 24 - November 6 2012 September 16 - October 23 2012 Emphasis on candidates' families, personal life, and 'human' moments Ad hominen attacks or substantial references to opponent's character or policy Reference to ordinary Americans across the country Reference to famous, non-political personalities

All or most of Tweet is written in Spanish. Evidence of symbols related to America, for example flags

27

iconography Babies and youth Large crowds Multiculturalism Rural setting Urban setting Small group

and banners On-screen appearance of babies, children or young people On-screen appearance of large crowds at rallies or conventions On-screen appearance of non-white Americans Primary setting is rural Primary setting is urban On screen appearance of a small group of supporters

Appendix B: Sample data B.1: Sampled tweets captured in Dedoose for November 5th, 2012, for both campaigns
Romney campaign:
MittRomney With your help, we will turn our country aroundand get America back on the path to prosperity. Please votetoday http://t.co/uYh0QcDS10:55 PM Nov 6th, 2012 from web MittRomney I am asking for your vote because I want tokeep America the hope of the earth. Find out where to vote:http://t.co/uYh0QcDS7:57 PM Nov 6th, 2012 from web MittRomney .@PaulRyanVP and I are offering real reformsfor a real recovery. Help us deliver it get out and vote todayhttp://t.co/uYh0QcDS5:38 PM Nov 6th, 2012 from web MittRomney A brighter future is out there waiting for us.Lets choose it today. http://t.co/D0TsMzjB2:21 PM Nov 6th, 2012 from web MittRomney We know we can bring this country back, butwe need your help. Find out where to vote today:http://t.co/uYh0QcDS11:07 AM Nov 6th, 2012 from web MittRomney If you are ready for real change, vote tomorrowfor the kind of leadership that these times demandhttp://t.co/uYh0QcDS12:08 AM Nov 6th, 2012 from web

Obama campaign:
BarackObama Time to vote! Grab a friend to join you andhead over to your polling place: http://t.co/Nq32LMcd,http://t.co/bgoYhI3v1:39 PM Nov 6th, 2012 from web BarackObama 25 reasons that 25 people are voting forPresident Obama: http://t.co/S1cAuY9u1:12 PM Nov 6th, 2012 from web BarackObama President Obama tells the story of "Fired up!Ready to go!" at the final rally of his final campaign.http://t.co/UqeF0v4g #Forward6:07 AM Nov 6th, 2012 from web BarackObama Iowa, tomorrow lets remind the world justwhy the United States of America is the greatest nation onEarth. President Obama4:37 AM Nov 6th, 2012 from web BarackObama But after a few minutes, Im feeling kind offired up. Im feeling kind of ready to go.4:32 AM Nov 6th, 2012 from web BarackObama "I got a powerful reminder of this myself onour last campaign."4:27 AM Nov 6th, 2012 from web

28

B.2: Sample of photos captured for analysis from Romney and Obama Facebook pages Obama campaign:

Romney campaign:

29

B.3: Sample of videos captured for analysis from Romney and Obama YouTube pages Obama campaign: (http://www.youtube.com/watch?v=trPn8sNOvYA)

Romney campaign: (http://www.youtube.com/watch?v=_REZmA3dGhA)

30

University of Oxford CENTRAL UNIVERSITY RESEARCH ETHICS COMMITTEE (CUREC) CUREC/1A Checklist for the Social Sciences and Humanities
The University of Oxford places a high value on the knowledge, expertise, and integrity of its members and their ability to conduct research to high standards of scholarship and ethics. The research ethics clearance procedures have been established to ensure that the University is meeting its obligations as a responsible institution. They start from the presumption that all members of the University will take their responsibilities and obligations seriously and will ensure that their research on human subjects is conducted according to the established principles and good practice in their fields and in accordance, where appropriate, with legal requirements. Since the requirements of research ethics review will vary from field to field and from project to project, the University accepts that different guidelines and procedures will be appropriate. Please check the CUREC website to ensure that you have the correct form for your project. This form does not cover research governance, satisfactory methodology, compliance with the requirements of publishers when administering their tests or questionnaires, or the health and safety of employees and students. As principal investigator, it is your responsibility to ensure that requirements in these areas are met. Please carry out a risk assessment of the project, in consultation with all researchers involved, using the checklist and CURECs other documentation. The use of an asterisk in this form indicates a phrase defined in the glossary. The glossary and further information on the Universitys research ethics procedures are available from t he CUREC website: www.admin.ox.ac.uk/curec This form is designed largely for research that falls within the Divisions of Social Sciences and Humanities and which does not involve a high level of risk to the subjects. Elite interviews, field work and oral history are included in the CUREC process. Please take a moment to read through it and if you have any questions or doubts as to whether it is the appropriate form, please review Section A or consult the CUREC website. Note on anonymised data and audit: you do not need to obtain ethical approval for your study if: you are using previously collected anonymised data about people which neither you nor anyone else involved in your study can trace back to the individuals who provided them (e.g. census data, administrative data, secondary analysis). Please refer to the definition of *personal data in the glossary and FAQ A4 for further guidance; or you are conducting research on behalf of or at the request of a service provider that matches the definition of *audit in the glossary.

If your research is audit or uses prior-anonymised data, please check this box: You do not need to seek ethical approval from CUREC, and you do not need to complete any more of this form. However, please check with your departments requirements, as some departments require you to lodge this form with them.

Office use only: IDREC Ref. No. __________________ Date of confirmation that checklist accepted on behalf of IDREC: // //

31

SECTION A 1) Are you using research methodologies commonly used in biomedical or behavioural laboratory sciences? 2) Is there a significant risk that the research will induce anxiety, stress or other harmful psychological states in participants that might persist beyond the duration of any test or interview in which they are participating? 3) Will the research involve human participants recruited by means of their status as present or past NHS patients or their relatives or carers? 4) Does the research involve *human participants aged 16 and over who do not have *capacity to consent for themselves? See the Mental Capacity Act 2005 5) Is the study to be funded by the US National Institutes of Health or another US federal funding agency? If you answered yes, please stop work on this checklist and for questions 1 and 2, complete CUREC/1 instead (available from www.admin.ox.ac.uk/curec/); for questions 3 and 4, submit your proposal to the appropriate NHS ethics committee (see www.nres.npsa.nhs.uk and www.admin.ox.ac.uk/researchsupport/ctrg for further information); for question 5, or if you answered 'yes' to questions 1, 2 or 4 and your research will take place outside the EU and is a biomedical study, submit your proposal to OXTREC, which uses separate documentation. Applications to OXTREC using this form will not be accepted. If your research is not a biomedical study and does not have US funding, but will take place outside the EU, you may use this form to submit your application for approval to the Social Sciences and Humanities IDREC. If you have answered no to all questions in Section A, please complete Sections B -E. This form and any supporting materials should be typewritten. SECTION B *Principal investigator/ supervisor/student researcher (title and name): Name of supervisor (STUDENT RESEARCH PROJECTS ONLY):

Yes

No X X

X X X

Josh Cowls

Dr Eric Meyer

Degree programme, e.g. DPhil, MSc Social Science of the Internet MPhil, MSc (STUDENT RESEARCH PROJECTS ONLY): Department or institute: Address for correspondence (if different): Email and phone contact: Title of research project: Josh.cowls@oii.ox.ac.uk 07969440591 Social Media and the 2012 US Presidential Election Oxford Internet Institute

32

Brief description of research methods and goals plus description of the nature of participants (including the criteria for inclusion/exclusion, method of recruitment), explanation of how professional guidelines and/or CUREC protocol(s) will be applied (if relevant) and expected use to which the results/data will be put. Please describe how you will obtain informed consent. Approx 400 words. The research methodology involves qualitative analysis of the tweets, Facebook posts and videos on YouTube sent out by the Obama and Romney campaigns during the course of the 2012 election. As these are all freely available to any user, there are no participants as such and thus no issues associated with recruitment or informed consent. The data will be analysed with a view to producing a detailed description of the social media strategies undertaken by the campaigns.

List actual or probable location(s) where project will be conducted, if known: Anticipated duration of project: Anticipated start date: Anticipated end date: Name and status (e.g. 3 year undergraduate; post-doctoral research assistant) of others taking part in the project:
rd

Online research

2 months 19 14 /01 /03 /2013 /2013

Please indicate what training on research ethics the researchers involved with this study have received, e.g. the title of the online or in-person course, and date completed (online training available at www.admin.ox.ac.uk/researchsupp ort/integrity/human/):

33

SECTION C Methods to be used in the study (tick as many as apply: this information will help the committee understand the nature of your research and may be used for audit). Please tick Interview Questionnaire Analysis of existing records Participant performs verbal/paper and pencil/computer based task Measurement/recording of motor behaviour Audio recording of participant Video recording or photography of participant Physiological recording from participant Participant observation Covert observation Systematic observation Observation of specific organisational practices Other (please specify) X

SECTION D Have you read one or more of the following professional guidelines and do you undertake to use the principles listed there as a guide for your own work? Please note that this is not intended to be an exhaustive list. Links to the guidelines listed below are included on the CUREC website. Please tick British Society of Criminology: Code of Ethics for Researchers in the Field of Criminology [www.britsoccrim.org/codeofethics.htm] British Educational Research Association Ethical Guidelines for Educational Research [www.bera.ac.uk/guidelines] Academy of Managements Code of Ethics [www.aomonline.org/aom.asp?ID=&page_ID=242] Association of American Geographers Statement on Professional Ethics [www.aag.org/cs/resolutions/ethics] Oral History Society of the UK Ethical Guidelines [www.oralhistory.org.uk/ethics/index.php] American Political Science Association (APSA) Guide to Professional Ethics in Political Science (Section H) [www.apsanet.org/content_9350.cfm] Political Studies Association Guide to Good Professional Conduct (see section on Research) [www.psa.ac.uk/Pubs] British Psychological Society Code of Ethics and Conduct [www.bps.org.uk/what-wedo/ethics-standards/ethics-standards] Ethics Guidelines of the Association of Social Anthropologists of the UK and Commonwealth [www.theasa.org/ethics/guidelines.shtml] Social Research Association: Ethical Guidelines [www.the-sra.org.uk/guidelines.htm] X

34

Statement of Principles of Ethical Research Practice from the Socio-Legal Studies Association [www.slsa.ac.uk/content/view/247/270/] Statement of Ethical Practice for the British Sociological Association [www.britsoc.co.uk/about/equality/statement-of-ethical-practice.aspx] Other professional guidelines (please specify):

SECTION E Please put a tick in the yes/no column as appropriate to indicate your response. 1) Will you obtain informed consent according to good practice in your discipline before participation? 2) Will you ensure that *personal data collected directly from participants or via a *third party is held and processed in accordance with the provisions of the Data Protection Act? 3) Does the research involve as participants *people whose ability to give free and informed consent is in question? (This includes those under 18 and vulnerable adults.) 4) As a consequence of taking part in the research, will participants be at serious risk of rendering themselves liable to criminal prosecution (e.g. by providing information on drug abuse or child abuse)? 5) Does the research involve the *deception of participants, as part of the investigation/experiment? Yes X Yes X Yes No X Yes No X Yes No X No No

If any of your answers above are in a shaded box, please indicate whether those aspects of your project are fully covered by the following. 6) Research protocol(s) which has/ve received IDREC/CUREC approval? If yes, please give protocol number(s): 7) Professional guidelines that you will be following, as noted under Section D? Yes No Yes No

If any of your answers in Section E are in a shaded box and are not covered by a protocol or by professional guidelines, please complete CUREC/2, available to download from the CUREC website. Then submit both this form (you need not complete section F) and the CUREC/2 to the Social Sciences and Humanities IDREC. If all your answers in Section E are in the unshaded boxes or your answers in shaded boxes are covered by a protocol or professional guidelines, complete Section F and submit this form and any accompanying documents to the Social Sciences and Humanities IDREC or to the relevant officer/committee at departmental level (see notes and address below).

35

FINAL CHECK Please check each of the following before submitting the checklist. If the appropriate supporting documentation is not included with your application, you will then be asked to provide this separately. This may well delay the ethical review process, and thus the start of your research. Have you completed Sections A-E? Have you defined all technical terms and abbreviations used? If you have produced any documentation in support of your application (which might include questionnaires, participant information, consent forms/form or note of procedure for recording oral consent, advertisements and surveys), have you attached a copy of these? Are all pages (including appendices and attachments) numbered? SECTION F You can submit this checklist by email and/or as a signed hard copy; if it is being sent by email only, the checklist, and any email from the head of department or nominee separately endorsing its submission, must be sent from a University of Oxford email address (i.e. as a minimum, the checklist and supporting documents must be submitted by the head of department or nominee indicating his/her approval from a University of Oxford email account). Complete this section only if you do not need to submit form CUREC/2. I understand my responsibilities as principal researcher/supervisor/student researcher as outlined in the CUREC glossary and guidance on the CUREC website. I declare that the answers above accurately describe my research as presently designed and that I will submit a new checklist should the design of my research change in a way which would alter any of the above responses so as to require completion of CUREC/2 (involving full scrutiny by an IDREC). I will inform the relevant IDREC if I cease to be the principal researcher on this project and supply the name and contact details of my successor if appropriate. Signed by principal researcher/supervisor/student researcher:Josh Cowls Date:28.01.2013 Print name (block capitals)JOSH COWLS Signed by supervisor:Eric T. Meyer(for student projects) Date:28 Jan 2013 Print name (block capitals)ERIC T MEYER I understand the questions and answers that have been entered above describing the research, and I will ensure that my practice in this research complies with these answers, subject to any modifications made by the principal researcher properly authorised by the CUREC system. Signed by associate/other researcher: . Print name (block capitals) Date X X

36

I have read the research project application named above. On the basis of the information available to me, I: (i) consider the principal researcher/supervisor/student researcher to be aware of her/his ethical responsibilities in regard to this research; (ii) consider that any ethical issues raised have been satisfactorily resolved or are covered by relevant professional guidelines and/or CUREC approved protocols, and that it is appropriate for the research to proceed without further formal ethical scrutiny at this stage (noting the principal researchers obligation to report should the design of the research change in a way which would alter any of the above responses so as to require completion of a CUREC/2 full application); (iii) am satisfied that the proposed project has been/will be subject to appropriate *peer review and is likely to contribute something useful to existing knowledge and/or to the education and training of the researcher(s) and that it is in the *public interest. (iv) [FOR DEPARTMENTS/FACULTIES WITH A DEPARTMENTAL RESEARCH ETHICS COMMITTEE (DREC) OR EQUIVALENT BODY - PLEASE DELETE IF NOT APPLICABLE] confirm that this checklist (and associated research outline) has been reviewed by the Departments Research Ethics Committee (DREC)/equivalent body, and attach the associated report from that body. Signed:.. (Head of department or nominee e.g Chair of DREC, Director of Graduate Studies for postgraduate student projects) Print name (block capitals) Date: If your research involves participants recruited by means of their status as current or former NHS staff, or the research will, in whole or in part, be carried out on NHS premises, use NHS facilities or assess NHS facilities or services, please see FAQ B3 (www.admin.ox.ac.uk/curec/faqs/). Please check with your department about its procedures for the approval of CUREC forms. If your department indicates that the checklist should be submitted directly to the IDREC, please send it, together with any supporting documentation, to the following address(es), keeping a copy for yourself: Secretary of the Social Sciences and Humanities IDREC ethics@socsci.ox.ac.uk University of Oxford Social Sciences Division Hayes House, George Street Oxford, OX1 2BQ Email:

IDRECs and/or CUREC will review a sample of completed checklists and may ask for further details of any project.

37

You might also like