Most people behave differently when they know they’re being watched, a fact that holds both in the real world and online. For many Internet users, the knowledge that their words and actions might be examined by the government leads them to self-censor opinions that they consider outside the mainstream.

That’s according to new research from Elizabeth Stoycheff, a journalism professor at Wayne State University. Last week, Stoycheff published a study in Journalism & Mass Communication Quarterly examining whether users would behave differently on social media if they were primed to think about government surveillance first.

To find out, she asked about 250 people to fill out a survey about their news consumption, social-media habits, and general attitudes toward government surveillance. After filling out the survey, participants viewed a fake Facebook news post about airstrikes against ISIS, and reported how willing they’d be to share their opinions on the subject. Finally, they were asked how they thought other Americans feel about the same topic.

A random subset of the survey respondents were shown a special notice before they saw the Facebook post, reminding them that the government “monitor[s] the online activities of individual citizens.”

Stoycheff found that people who said they think government surveillance is justified—about two-thirds of the respondents—were likeliest to alter their behavior after they were reminded about government surveillance. Specifically, these pro-surveillance Internet users tended to avoid sharing opinions that they believed were outside the mainstream.

One of the ways the study tested attitudes about surveillance was by asking people if they agreed with the statement, “The government can track my online behavior because I have nothing to hide.” But in fact, the people who agreed that they had “nothing to hide” were the people who were most likely to censor themselves.

Stoycheff says that attitude aggravates her. “When I talk to people about what I do and what I study, more often than not I get a response of, ‘Oh, it’s this inevitability that the government is going to monitor us. … And it’s not really affecting me because I don’t break the law.’ I find that such a problematic thing for people to say.”

By contrast, the smaller group of people who said that government surveillance is not justified took one of two routes: Either they refrained from posting their opinions on social media entirely, or they spoke their minds online, regardless of whether they thought they were expressing majority or minority opinions.

Stoycheff thinks that’s because people who are vehemently anti-surveillance tend to be educated, politically active, and willing to share controversial thoughts—or they’re so wary of surveillance that they don’t post about politics at all.

It’s not surprising that for most, a simple reminder of everyday surveillance was enough to alter people’s behavior. Studies have shown that being watched can make people less likely to commit some crimes, or to act in ways that don’t conform to widely held morals.

One 2013 study, for example, found that when restaurants installed monitoring software to alert managers about stealing employees, weekly revenue went up, suggesting that employees were changing their behavior in response to the knowledge that they were being watched.

In the case of the restaurants, surveillance helped guide people toward moral decision-making, because people were less willing to steal if they knew others were watching. But in the same way, being watched can move people toward a majority opinion rather than toward moral actions.

In real life, of course, Facebook won’t show you a jarring reminder of NSA surveillance every time you post a status about a sensitive topic. But those reminders exist in the wild, Stoycheff says, in the form of news stories about surveillance, and the ubiquitous user agreements that services force you to accept before signing up—each an example of the many ways that your data is captured and used.

And the silencing effects of surveillance are especially harmful for groups of people who generally consider their opinions to be outside the majority.

Before we spoke on the phone this week, Stoycheff reexamined her data to see if members of different racial groups behaved in different ways. She found that the non-white respondents in her study (who made up 26 percent of participants) were more likely to say that they did not hold majority opinions.

That fact suggests that people of color are more likely to suppress their non-majority opinions, as long as they consider government surveillance justified. (The distribution of pro- and anti-surveillance people among non-whites was about the same as it was among whites, with two-thirds in support of surveillance.)

These results show that surveillance can create something of an echo chamber, amplifying widely-held opinions and weeding out other perspectives. Surveillance “changes the assumption that we’ve been working on this whole time, that the Internet is a safe space for deliberation, for people to share their ideas,” Stoycheff said. “All of a sudden, that may not be the case.”