Varying Bits: A Computational Perspective on News Diversity and Political Parallelism

Abstract

News media play a pivotal role in the functioning of democracies.They facilitate information exchange between elected officials and the public, have the capacity to mobilize social groups and can provide interpretation and context to the events that take place in the world around us. Considering these roles of news media, my goal with this thesis is to investigate how newspapers in Norway, The Netherlands, Denmark and the United Kingdom fulfill these roles, specifically by looking at how diverse and politically slanted their news coverage is. I do this during a period (2000-2020) in which the Internet is providing ever increasing competition for regular news media, such as newspapers. By utilizing computational text analysis methods, it is possible to consider every news article from the newspapers included in this study, and conduct analyses that span a long period of time. Through these analyses, I aim to contribute to the empirical knowledge on news diversity and political parallelism, and to investigate and improve upon the computational methods that are available to measure these concepts. Concretely, the contributions in this thesis are structured around three studies. The first is focused on the development/improvement of a method to generate sentiment dictionaries, so that it can be used for evaluating valence in political news articles. The second study investigates the diversity between newspapers in terms of lexical and valence diversity, while the third evaluates the presence of bias in the amount and valence of attention for specific political parties.

The results indicate a modest increase in news diversity, rather than the expected theory-based decline. Political parallelism either remains stable or slightly decreases, depending on the country. Both of these findings can be considered positive from a normative standpoint. Strong trends in news diversity might be detrimental to the roles that news media fulfill in democracies, resulting in either over or under-representation of specific issues. Similarly, newspaper readers are not exposed to increasingly slanted political news, as political parallelism remains stable. However, the amount of political parallelism is substantial, indicating that readers of different newspapers are exposed to different kinds of political slant. As for valence, the results show that there is no substantial difference between newspapers in the valence with which they discuss the same events, nor is there any political parallelism in the valence of political party coverage. The contribution of these empirical findings is twofold. First, the findings illustrate that logical and plausible theoretical assumptions do not always properly reflect the reality on which they are based. Hence, computational methods are shown to be capable of providing new and possibly surprising perspectives. The second contribution is then found in detailing the ways in which computational methods can contribute to the ongoing academic discussion on long-standing theories and assumptions.