Chinese censors are more worried about online discussion of protests and other "collective action events" than posts on senior leaders, government policies and other sensitive issues, research by Harvard University shows.
The findings contradict the widespread belief that internet censorship is designed to target sensitive words, including the names of state leaders.
Even posts praising the central government could be censored if they talk about such events, the researchers found.
"Collective actions", as defined in the survey, are events that organise crowds and relate to individuals who have initiated protests and relate to nationalist sentiments used to encourage people to take to the street.
Posts that mention an ongoing collective action event are 20 to 40 per cent more likely to be censored than those involving non-collective action events, such as a discussion of President Xi Jinping, corrupt officials, government policies and issues such as Tibet and Xinjiang . The findings of the study, by Professor Gary King of Harvard's government department and two postgraduate students, are published in the latest edition of Science magazine.
Mara Hvistendahl, a contributing editor of Science who is based in Shanghai, said: "It makes sense because collective action events are more threatening to both local and central government. One of the biggest fears about social networks in China and other countries is that it could be used to mobilise people."
The researchers created two separate accounts on each of 100 Chinese social media platforms and online forums, including Weibo. They published 1,200 posts. Some contained discussions of collective actions, such as a protest because of inadequate compensation for the seizure of collectively owned farmland. Others simply touched on sensitive words including "Tibet", "Xinjiang" and names of senior leaders.
Within the same time period, posts about collective action events were more than 30 per cent more likely to be censored than the posts about non-collective events.
"Criticism of the state is quite useful for the government in identifying public sentiment, whereas the spread of collective action is potentially very damaging," Margaret Roberts, one of the postgraduate students, said in a separate review.
The finding echoes the team's previous research on Chinese internet censorship, in which they collected more than three million posts before they were reviewed by the authorities and revisited to see which ones were censored.
Over half of the posts about collective action, no matter if they were criticising or supporting the authorities, were deleted. But only around 24 per cent of the posts on what the researchers classified as sensitive topics, including criticism of state leaders, got deleted.
The later research also revealed that computer programs that scan posts and delete those with sensitive key words before they are published online were far less effective than internet monitors who delete posts after they appeared online.
“The study provides convincing evidence for our understanding of the internet censorship decision-making mechanism used by Chinese authorities or the Chinese social media companies,” said Dr Fu King-wa, an assistant professor at the University of Hong Kong who studies Weibo censorship and led the development of Weiboscope, a program that recovers deleted posts.
But the Harvard research did not reveal the full picture of the mechanism, Fu said. “Images were not part of the study. But the majority of our collected censored posts had trivial text contents. They were attached to images instead,” he said.
Fu added that timing is also a decisive factor in the censors’ decision making. “Recently [after the investigation of several state television executives], posts about a Central China Television presenter were censored. We did not see that happening before then.”