Understanding Election Twitter
What Internet search trends do and do not tell us about politics
July 16, 2009
This summer's major elections in India, Iran, and Indonesia have happened in an increasingly online world where Internet penetration has grown to significant proportions. Yet despite all the novel platforms for search, debate, and micro-broadcast, the extent to which social scientists can use the Internet to understand political trends remains unclear. For example, Internet penetration may be low in India, but the aggregate number of users is equivalent to usage in Japan—around 80 million. In the Middle East, the Iranian election highlighted the impact 23 million people could have by mobilizing via Twitter.
In Indonesia, while candidates politicized the jilbab, or head scarf, Google search records indicate that the primary online interest was for fashion. Over recent months, despite media hype surrounding the impact of candidate debate on the role of Islam in Indonesia, Google queries for "jilbab" declined by 20 percent. While relative search on the term did spike periodically, including after the final debates on July 2, searches for "Jilbab Paris" increased by 60 percent in Jakarta. In this case, polling and Internet search analytics indicate divergent realities—on the politicization of jilbab, polling indicates substance while Internet search analytics point to style.
The Internet has created what E. M. Forster might have called Screens with a View. Blogs overview local issues and cohere around central issues according to common patterns of linking. Aggregation sites such as Twitter make remote syndication simple and convenient, allowing private thoughts to become public chatter. Platforms such as Facebook stretch communities and add depth to conversations through features such as News Feed, which allows weakly bonded individuals to see each other's life updates. Facebook conversations and Google queries can be aggregated and viewed as relative longitudinal trends, and Google's Insights for Search displays user interests across geography. These cutting edge tools will impact the way politics is run from the inside and observed from the outside.
Research is emerging on this new social and political topography. For example, a recent map of the Arabic Blogosphere from Harvard University's Berkman Center for Internet and Society analyzes the content and linking patterns of 35,000 Arabic-language blogs across 18 countries in the Middle East, developing a modern sociogram of Internet relationships. Authors Bruce Etling, John Kelly, Rob Faris, and John Palfrey created a three-dimensional map of 12 topical clusters across Middle Eastern blogs, and revealed the importance of local, rather than controversial, issues in such public-public platforms. It demonstrated that an exploding blogosphere is widely exposing pluralistic ideas across political boundaries, and linguistic barriers are falling due to "bridge bloggers" who cross the English-Arabic divide in the Levant and the French-Arabic divide in the Maghreb.
Understanding what people are reading and searching is equally important. Through Google Insights for Search, one can compare user queries across time and geography. Such relative data, maintained in the aptly named Google Zeitgeist, reveals the spirit of the times.
Google Insights for Search cannot hope to become a crystal ball, and no country's voting patterns can be summoned from the Internet data, but relative search data is useful insofar as it acts as a leading indicator of user interest. Internet search engine query data reveals online curiosities that are empirically translating into real-world actions. As such, grounded understanding of online trending is useful in making real-world predictions. For example, Google.org—the philanthropic arm of Google that invests in innovative green energy and health technology—has partnered with the Centers for Disease Control (CDC). Findings published in a November 2008 Nature article indicate that online behavior such as search volume on key terms such as "flu" acts as a leading indicator of influenza activity across U.S. cities. This collaborative partnership has allowed the CDC to quickly patrol and minimize outbreaks. The sharing of anonymous data neither predicts nor prevents epidemics, but it accelerates response time, facilitates containment, and has likely abated isolated cases of flu proliferation.
Online political information-seeking is similarly illustrative, if not yet predictive. It did not tell us that in India the Congress Party would obtain more Lok Sabha seats in May 2008. Nor did it predict that Ahmadinejad would be reelected over Mousavi in Iran. Nor did it decisively tell us that incumbent Susilo Bambang Yudhoyono would be reelected in Indonesia's second democratic presidential election.
Information gathering—whether via newspaper, blog, search, or short message service—is indicative of electoral interest, but not necessarily an ability to influence its course. Internet search analytics indicate volume, but not necessarily a disaggregation between positive and negative perception. For example, while pre-election query volume for Indonesian opposition party candidate Megawati appeared initially comparable to that for incumbent Yudhoyono, closer inspection indicated rising popularity on the term "say no Megawati." As the adage states, all publicity may be good publicity, but not all search queries are created equal—some reveal opposition as much as support. Google Insights for Search offers related search terms to qualitatively contextualize trends. Volume must be coupled with a related term to reveal the trend's amplitude and direction.
In evaluating the extent to which relative trends in Internet search data are predictive, one must be cautious to first understand the domestic levers of political influence. Relative Internet search volume data may yet prove predictive in isolating demographic or geographic voting patterns, but only when users, access to information, and breadth of discourse are understood, and only when being informed translates to being influential.
The bigger question is, How does what happens online translate into pressure on leadership, and influence policy outcomes? Understanding the leading interests of particular demographics through search trending and social network perspectives will facilitate an understanding of the offline implications.
Scott E. Hartley is a researcher at Harvard University's Berkman Center for Internet & Society, and a joint-degree graduate student at Columbia University.