Detecting coverage bias in user-generated content

Abstract

ABSTRACT The importance of user-generated content is growing as media consumption is moving online; yet, investigations of media bias on user-generated content platforms are rare. We develop a novel procedure to detect coverage bias – i.e., bias in the amount of coverage certain topics or issues receive – on user-generated content platforms. We proceed in two steps. First, we focus on a sample of homogeneous observations and control for observable differences. Second, we compare the coverage of our observations between different language versions of the same platform in a difference-in-differences framework, which allows us to disentangle coverage bias from unobserved heterogeneity between observations. We apply our procedure to Wikipedia and examine whether it has a coverage bias in its biographies of German (and French) Members of Parliament (MPs). Our analysis reveals a small to medium size coverage bias against MPs from the center-left parties in Germany and in France. A plausible explanation are partisan contributions to the Wikipedia biographies, as we show by analyzing patterns of authorship and Wikipedia’s talk pages for the German case. Practical implications of our results include raising users’ awareness of coverage bias when searching for and processing information obtained on user-generated content platforms.