How accurate is media-elicited protest event data from autocracies where the state censors the media? Based on a source-specific model of event selection and a multisource dataset of over 3,100 protests from three Chinese megacities, we demonstrate the substantial advantages of using social media data, capturing 115 times more protests than English-language international news, 74 times more than domestic news, and 10 times more than dissident websites. Social media are most likely to cover small and nonviolent events that other sources often ignore. Aside from antiregime protests, they are less affected by censorship than often assumed. A validity test against public holidays and daily rainfall shows that social media data outperform dissident websites and traditional news. Social media, and to a lesser extent dissident media, are promising new sources for protest event analysis in autocracies. Scholars should treat news media-based event data from heavily censoring regimes with caution.

This content is only available as a PDF.

Author notes

* The research for this article was funded by the Hong Kong Research Grants Council, Early Career Scheme (Grant No. 24615215), the European Research Council (Grant No. 678266), and the Chiang Ching-kuo Foundation for International Scholarly Exchange (Grant No. CS002-U-18). The authors would like to thank Anita Gohdes, Swen Hutter, Tianguang Meng, Zachary Steinert-Threlkeld, Yang Su, Nils Weidmann, and five anonymous reviewers. They provided very valuable feedback on previous drafts of this article. The authors dedicate this article to Lu Yuyu and Li Tingyu for their path-breaking contribution to science.