Political Communication on Social Media in the Run-Up to the 2013 German Federal Election

The goal of this project was to explore political communication on social media in the run-up to the 2013 German federal election. The empirical analyses focused on the micro-blogging service Twitter. For data collection, we employed Twitter’s official data vendor Gnip. To establish an initial dataset covering politically relevant messages posted during the campaign for the German federal election 2013, we queried the Gnip Historical Powertrack for messages containing the names of political parties and candidates, campaign-related phrases, and keywords related to campaign-related media events. For the analyses, we selected the messages with a sufficiently high likelihood of referring to German politics. This data preparation yielded a data set comprising almost 1.4 million messages from roughly 100,000 users.

For the analyses, we built on a model of data-generating processes underlying aggregate statistics of digital trace data. This model spells out conditions under which aggregate statistics of Twitter conversations may serve as valuable indicators of offline phenomena, thereby casting some doubts on the general utility of these statistics for this purpose. The empirical analyses focused on selected claims about the utility of digital trace data. In one line of research, we addressed the relationship between Twitter conversations and political reality. Using data from the 2013 German federal election campaign, we demonstrated that Twitter conversations did not provide an unbiased portrayal of political events, popular topics of discussion, and attention toward political actors. Instead of providing a true image of political reality, the evidence is in line with a model suggesting that the interplay of factors influence the mediation of political reality through Twitter. Another line of research addressed the capability of aggregate statistics of Twitter conversations to reflect the current popularity of political parties and predict the outcome of elections. Building on a theoretical analysis, we found little, if any, support for the idea that Twitter-based metrics are a valuable indicator of political support over the course of the campaign for the 2013 federal election. Instead, the analysis provided considerable evidence that Twitter-based metrics reflect users’ attention to politics, which may be considered a covariate of political support in specific cases, but not in any circumstances. Thus, Twitter may have the potential to become a source of insight into attention toward politics rather than an indicator of political support.

Taken together, the project examined popular ideas about the general utility of digital trace data as indicators of political offline phenomena from both a theoretical and an empirical perspective. In effect, it helped to create more adequate ideas of the conditional utility of digital trace data as indicators of offline phenomena.