Political media bias is an important problem in political communications literature. In this article, extending on the selective exposure theory by publicly available social media data, I introduce methods for measuring media-party parallelism (MPP) and exploit quasi-experimental designs (QEDs) to understand some of its dynamics. To evaluate the robustness of the methods better, a country whose media system is known to have a high degree of polarized pluralism and political parallelism is chosen as a test case. For that matter, two kinds of observational datasets of Turkey is compiled using Twitter API: the first being the news audience dataset and the second one relating to the political audience. By proposing a similarity metric for news organizations and adopting a network theoretical approach, I first provide a computational framework for inferring media groupings at different granularities. Then, to quantify the MPP, I utilize two indicators and adapt them to the social media context: effective number of parties (ENP) and group descriptiveness (GD) of the media. Furthermore, I exploited QEDs to test the hypothesized effects of two major political developments in Turkey on the media landscape. Findings regarding MPP are not only verified by comparing the online audience compositions before and after the major developments as a longitudinal study, but also validated by off-line cross-sectional surveys.
Code is available on GitHub