Online anonymity: 'Stable pseudonyms' create a more civil environment than real user names 

The ability to remain anonymous when commenting online is a double-edged sword. It is valuable because it enables people to speak without fear of social and legal discrimination. But this is also what makes it dangerous. Someone from a repressive religious community can use anonymity to talk about their sexuality, for example. But someone else can use anonymity to hurl abuse at them with impunity.

Many people focus on the dangers of online anonymity. Back in 2011, Randi Zuckerberg, sister of Mark and (then) marketing director of Facebook, said that for safety's sake, "anonymity on the internet has to go away". Such calls appear again and again. Behind them is a common intuition: that debate would be more civil and constructive if people used their real names.

But my research with colleagues suggests that anonymity—under certain conditions—can actually make for more civil and productive online discussion. This surprising result came out of a study looking at the deliberative quality of comments on online news articles under a range of different identity rules.

We built a data set of 45 million comments on news articles on the Huffington Post website between January 2013 and February 2015. During this period, the site moved from a regime of easy anonymity to registered pseudonyms and finally to outsourcing their comments to Facebook. This created three distinct phases.

In the initial phase users could easily set up multiple accounts. The comment space was, at that time, a troll's paradise. People could read an article, quickly create a username, and post whatever they wanted. If moderators blocked that username for abusive behavior, the person (or even bot) behind it could just make another, and then another, and so on. This led to a space that was unpleasant for users. So the website began to make changes.

In the second phase, users had to authenticate their accounts, but did not have to use their real name with their comments. That meant they could be anonymous to other users but could be identified by the platform. If they behaved badly and were blocked, they couldn't just make a new account and carry on—at least, not without creating a new authenticating account on Facebook. This made personas on this commenting space less disposable. They became "stable pseudonyms."

In the third phase, the commenting system was outsourced to Facebook. Huffington Post usernames were replaced with user's Facebook names and avatars. Depending on settings, comments might appear on users' Facebook feeds. While not everyone has their own face on their profile picture, and not everyone even uses their real name on their account, many users do. This third phase therefore roughly approximates a real-name environment.

Keeping it friendly

We looked initially at the use of swear words and offensive terms—a crude measure of civility. We found that after the first change the use of these words dropped significantly. This was not just because some of the worst offenders left the site. Among those who stayed, language was cleaner after the change than before. We describe this as a sort of "broken-windows" effect, after the famous theory that cleaning up a neighborhood can help reduce crime. Here, a cleaner environment improves everyone's behavior.

We then looked across all three phases at other features of individual comments, including the length of words, causation words (for example, "because"), words indicating tentative conclusions (for example, "perhaps"), and more. We were able to automate this analysis and use it to construct a measure of the "cognitive complexity" of comments. This method has been tested on the deliberations of the Swiss parliament and shown to be a good proxy for deliberative quality. We could not, of course, see the context and meaning of each individual comment, but using this method at least allowed us to do the analysis at a very large scale.

Our results suggest that the quality of comments was highest in the middle phase. There was a great improvement after the shift from easy or disposable anonymity to what we call "durable pseudonyms." But instead of improving further after the shift to the real-name phase, the quality of comments actually got worse—not as bad as in the first phase, but still worse by our measure.

A surprise finding

This complicates the common assumption that people behave better with their real names on display. We don't know exactly what explains our results, but one possibility is that under durable pseudonyms the users orient their comments primarily at their fellow commentators as an audience. They then perhaps develop a concern for their own reputation within that forum, as has been suggested elsewhere. It's possible that a real-name environment shifts the dynamic. When you make comments that can be seen not only by other Huffington Post readers but also by your Facebook friends, it seems plausible that you might speak differently.

What matters, it seems, is not so much whether you are commenting anonymously, but whether you are invested in your persona and accountable for its behavior in that particular forum. There seems to be value in enabling people to speak on forums without their comments being connected, via their real names, to other contexts. The online comment management company Disqus, in a similar vein, found that comments made under conditions of durable pseudonymity were rated by other users as having the highest quality.

There is obviously more to online discussion spaces than just their identity rules. But we can at least say that calls to end anonymity online by forcing people to reveal their real identities might not have the effects people expect—even if it appears to be the most obvious answer.