Chinese Social Media Censorship May Be More Selective Than Previously Thought

A new study finds that Chinese censors target potential calls to action rather than anti-government criticism.

The Chinese government may be much more selective about what it censors than previously thought, according to a new study, by Harvard government professor Gary King, Harvard grad student Jen Pan and Harvard Ph.D candidate Molly Roberts, published today by

It's undisputed — both inside the country and around the world — that the Chinese government censors free speech, on the Internet and elsewhere. Until now though, most have accepted the theory that Chinese leaders actively prune the Internet of anti-government or critical commentary and posts. However, after analyzing tens of millions of Chinese social media posts and even launching their own Chinese social media platform, the study's authors aren't convinced that Chinese censorship is that simple.

"The [current] theory was that the next person would come to the web and would only see more positive things than what the reality represents," King said. "But think about if you're in a country where you know there's censorship and you only see positive things about the government...would that make you think more positive things or negative things?"

Chinese leaders, Professor King said, are mostly interested in preserving their positions of power. The biggest threat to that claim to power are the Chinese people, according to King. And as it happens the best way for a group of people to incite change, or say, attempt to topple an existing regime, is to promote collective action. "The thing they have to worry about is something like Tiananmen Square," King said.

So rather than focus on weeding out every iota of negativity and criticism that exists on social media sites, the Chinese government focuses on silencing people or posts that ask others to mobilize regardless of whether it's for a pro- or anti-government purpose.

After conducting a study of what originally began as 11 million social media posts, King, Pan and Roberts found that criticism on social media is useful for the Chinese government so long as that criticism does not include a call to action. In an effort to mollify Chinese citizens and essentially prevent any form of collective action, the Chinese government will take any negative commentary on policies or leaders into consideration and make the changes they see fit.

The difference between a citizen who turns to social media to simply air his or her grievances about the government and a pro-government citizen who is attempting to mobilize a large group of people is that the latter holds a power that in the long run could be potentially dangerous to the government's regime and is therefore prioritized for censorship.

The study's conclusion — that the government's social media censorship does not target criticism but rather potential calls to action — might bring about some skepticism among those who either are subject to Chinese censorship or have examined the censorship practices employed by the government. But a study of Chinese censorship at this scale has yet to be done, King said.

The experiment began as King, Pan and Roberts were initially developing ways to analyze large masses of text. After honing their methods in English, the trio looked to stress test the experiment and see if it worked in multiple situations and languages. So they turned to Crimson Hexagon, a social media analytics firm at Harvard, which gave them 11 million Chinese-language social media posts to test their analysis methods on. When Pan and Roberts — both of whom speak and read Chinese fluently — tried to access the posts, they found that some of the URLs lead to either a dead link or a site that indicated that particular post had been flagged for review. Pan and Roberts realized they were sitting on a database of posts that were still in the review stage of the censorship process.

In order to understand the mechanics of the censorship process of social media sites, they bought a Chinese URL and created a social media website. Though only the three of them would post to the site, King was not only given documentation that explained how to properly run a social media site according to the Chinese government but was also able to freely communicate with customer service representatives who were trained specifically to answer questions about how to stay out of trouble with the Chinese government.

From the customer service representatives, King, Pan and Roberts learned that there are are two ways that the Chinese government reviews posts: human labor and an automated keyword review process. According to King, China has 100,000 to 200,000 people trained and employed as censors. They painstakingly conduct manual reviews to see if these posts either incite collective action or otherwise meet qualifications to be censored. First, a keyword matching process flags suspect posts and directs it into a massive queue, where a human censor decides if the post should or shouldn't be published. In other cases, a post will be automatically published and a human censor will review it within 24 hours of its publication.

Review posts in hand, King, Pan and Roberts had the ability to track these many posts and see which of these would be censored. Given the sheer scale of the data they had in their hands, patterns quickly emerged.

Next, King, Pan and Roberts created 200 accounts on Chinese social media sites. They wrote some posts that called for collective action, some in support of the government, some against and waited to see if the posts "came out of the other side."

"We went back to each post from our network of computers around the world to see which ones were censored," King told BuzzFeed.

According to the article they found that, "66 of the 100 sites...(automatically) review at least some social media submissions, and 40% of all of [the] individual social media sub- missions from...100 sites (and 52% of submis- sions from sites that review at least sometimes) are put into review. Of those submissions that go into review, 63% never appear on the web."

The biggest takeaway of this experiment, King said, was the paradoxical nature of censorship in China that only becomes apparent when examining the practices from afar and at scale.

"The goal is to stop the flow of information but they also [use social media to] convey a lot about the intentions of the government," he said. "If you're a leader of China you have to make sure there' real event that involves collective action. You have to stop discussion of it. If there's criticism on the web you can replace leaders before there's collective action."

Put simply, the Chinese government seems less concerned with what its citizens say and is far more invested in moderating what they do.