Topic modelling as a strategic campaigning tool

07.03.2018

Topic modelling is a qualitative and quantitative method of identifying structures in texts. It is an increasingly common part of the campaign monitoring process. The applicability of topic modelling exemplified by the third corporate tax reform.

Topic modelling is spread alongside online communication, as it effectively condenses the mass of information that is made machine-readable. Its use in political campaigns, such as referendums, has been increasing steadily as a result.

Case study: corporate tax reform

In early 2017, the Swiss people voted on reforming the country’s corporate tax system (USR III). Applying topic modelling to the online articles on the third corporate tax reform yielded six to eight topics dominating the debate leading up to the referendum. In late November 2016, the gfs.bern research institute identified those topics anew every week. Its findings were astounding: even though the articles changed on a weekly basis, the fundamental structure of the topics remained similar at least. This made it easy to determine the actual key issues of the ongoing controversy.

Reading aid: Each dot is an online article. Each colour corresponds to a topic. The larger the circle, the clearer its association with the topic. Connecting lines indicate relations within a topic. Thicker lines connect more similar articles.

Between December 2016 and January 2017, the seven topics of the debate around the third corporate tax reform were:

  • Economic benefits
  • Controversy in Basel
  • Controversy in Zurich
  • Meta-communication about the referendum
  • Tax losses for communities and cities
  • Church resistance
  • Tax losses for the Federal Government

Which controversies were discussed with particular intensity in the context of the seven topics listed above? 

Topic 1: Economic benefits

Economic benefits were the most frequently presented argument in favour of the proposed reform. Its proponents emphasised that the reform would secure Switzerland’s status as an attractive business location. Other key words included research and jobs. Views in favour of the reform dominated this topic.

Topics 2 and 3: Regional controversies

The regional controversies in the Basel and Zurich areas were frequently covered in the media. This topic focused primarily on the consequences of the reform for Winterthur and the canton of Basel-Landschaft. In the Basel region, the discussion was heavily personalised: at first, it was dominated by the councillor Eva Herzog, later, the position of Claude Janiak of the Council of States became more relevant to the debate. In Zurich, the tension between the canton and its towns/cities and local authorities was a talking point. Neither local topic was clearly dominated by arguments in favour of or against the referendum.

Topic 4: Meta-communication

In this topic, too, both sides of the debate were represented more or less equally. It comprised articles about the referendum as such and its campaigns. The surveys about the reform received considerable attention, with an emphasis on the openness of the decision.

Topic 5: Tax losses in local authorities and towns/cities

Opponents of the reform frequently cited its consequences for local authorities and towns/cities as an argument. They warned of empty treasuries and the expected impact on the middle classes.

Topic 6: Church resistance

The controversy surrounding resistance from the church (and the reactions from business leaders) were fairly ambivalent. This topic focused on social services provided by the national churches.

Topic 7: Tax losses for the Federal Government

The initially dominant controversy surrounding the consequences of the reform on the Federal Government received less attention over time. It focused on tax losses from the outset. A study conducted by the unions helped shift public focus to the billions that would be missing in the treasuries.

The key issue

The use of topic modelling allowed us to observe how the debate morphed from a discussion about the necessity of the proposed reform for the Swiss economy into a discussion about its damaging impact on individual communities. During the debate, the former argument increasingly lost ground, while proponents of the reform failed to tie their own arguments up with the discussion on the harm it would inflict on local communities. As a result, proponents of the reform were gradually excluded from the main debate and lost voter support in the process.

Topic modelling enabled us to identify prototypical articles within each topic. They exemplified the development of the debate.

A brief summary
Topic modelling has become an established method of analysis for political communication, especially in the online sphere. It introduces structure to mass information and quantifies typical communication structures. As a self-learning mechanised process, it can handle any volume of information and precisely detect the overarching storyline among the vast noise of the online and offline media.

Topic modelling rapidly identifies points where players can introduce their own topics into the core of the debate and points where their concerns would be swept away by the greater discussion. Further, the analysis systematically provides insights on which discussion topics are interlinked and, as such, support or contradict one’s own position.

In the context of a strategic campaign, it is an efficient way of identifying trends in the mass-media dissemination of topics.

The method is one of the latest tools used by gfs.bern to monitor political communication.