Swedish Public Service: Politically biased or not?- Using text mining on political speeches to measure the political content in public service news articles

Abstract: Public service in Sweden is supposed to be a politically independent source of information but in recent years this has been questioned due to its perceived left leaning by the population. In this paper the notion of a politically biased public service is studied following the approach by Laver et al (2003). A random forest approach was also proposed but with poor model performance. Parliament speeches between 2016 and 2021 were used to identify words and phrases with a political leaning to later use on article texts. Using the Laver method, a significant right bias can be found in most studied periods and the method performs well on positioning other news outlets relative to each other on a left-to-right dimension. The relative position of the public service SVT is to the right of all commercial news outlets, again suggesting a right bias in their news reporting. Further evaluation of the method on party speeches shows less convincing results, as all left speeches are calculated to be biased to the right. The reliability of the results should therefore be considered low.