Usually, we behave better when we know we’re being watched. According to decades of research, the presence of other people, cameras or even just a picture of eyes seems to nudge us toward civility: We become more likely to give to charity, for example, and less likely to speed, steal or take more than our fair share of candy.

But what happens when the cameras are on the chests of police officers? The results of the largest, most rigorous study of police body cameras in the United States came out Friday morning, and they are surprising both police officers and researchers.

For seven months, just over a thousand Washington, D.C., police officers were randomly assigned cameras — and another thousand were not. Researchers tracked use-of-force incidents, civilian complaints, charging decisions and other outcomes to see if the cameras changed behavior. But on every metric, the effects were too small to be statistically significant. Officers with cameras used force and faced civilian complaints at about the same rates as officers without cameras.

“These results suggest we should recalibrate our expectations” of cameras’ ability to make a “large-scale behavioral change in policing, particularly in contexts similar to Washington, D.C.,” concluded the study, which was led by David Yokum at the Lab @ DC, a team of scientists embedded in city government, and Anita Ravishankar at Washington’s Metropolitan Police Department.

After the public uprising in response to the 2014 police shooting of Michael Brown in Ferguson, Missouri, advocates and many police officials turned to cameras as a way to reduce violent encounters and build trust. By 2015, 95 percent of large police departments reported they were using body cameras or had committed to doing so in the near future, according to a national survey.

The cameras provide an independent, if sometimes ambiguous, record of police-civilian encounters.

Until now, the most commonly cited study on police body cameras had suggested that cameras did indeed have a calming effect. That experiment took place in 2012 in Rialto, California, where officers were randomly assigned cameras based on their shifts. Over a year, shifts that included cameras experienced half as many use-of-force incidents (including the use of a police baton, Taser or gun) as those shifts without cameras. The number of complaints filed by civilians against officers also declined — a stunning 90 percent compared with the previous year.

The Rialto study had a big impact in policing. Axon (formerly known as Taser International) has sold more than 300,000 police cameras worldwide and cites the Rialto study on its website. A federal district judge also cited the study in 2013 when she ordered the New York City Police Department to conduct a yearlong pilot program using body cameras. (Results are due out this spring.)

But the Rialto experiment featured just 54 officers, compared with over 2,000 in Washington. Officers in Washington captured five times as many hours of video. The larger sample size and the long-term way the cameras were assigned added to the reliability of the Washington results.

“This is the most important empirical study on the impact of police body-worn cameras to date,” said Harlan Yu from Upturn, a Washington nonprofit consulting company that studies how technology affects social issues. It was not directly involved in the research. “The results call into question whether police departments should be adopting body-worn cameras, given their high cost.”

The federal government has given police departments more than $40 million to invest in body cameras, and state and local authorities have spent many millions more. The devices vary in price, but the biggest expense is the data-storage cost. In Washington, MPD officers collect about a thousand hours of footage a day. About 40 percent of it is deleted within 90 days, while the rest is to be kept for months, years or decades, depending on the statute of limitations for the charges connected to the footage.

The cameras also have a cost in terms of privacy violations. In a report on the policies governing police body cams in 50 major departments, Upturn and the Leadership Conference on Civil and Human Rights found that many cities have weak rules in place. Those rules can allow departments to refuse to share footage with civilians who want to file complaints, for example.

Some companies are exploring ways to integrate facial recognition software into police cameras, a level of surveillance that would disproportionately affect low-income minority communities, where the police tend to spend the most time.

“D.C. has relatively strong body-worn-camera procedures,” Yu said. “In cities that have weaker policies, it’s certainly possible that the cameras may be doing more harm than good.” Like other advocacy groups, the has supported body cameras. But the study, says Monica Hopkins-Maxwell at the ACLU of Washington, “should give us pause.”

For now, Washington officials say they intend to continue using cameras. But they admit to being surprised by the findings.

“I thought it would make a difference on police and civilian behavior — particularly for officers, and this is the exception, who might be more inclined to misbehave,” said MPD Chief Peter Newsham.

Why didn’t the cameras change behavior? After all, we know that other cameras do change behavior. Public closed-circuit TV cameras seem to lead to a moderate decrease in crime, particularly in parking garages. Traffic cameras significantly decrease speeding and fatal accidents.

Even the suggestion that someone is watching us tends to influence us: In 2011, researchers at Newcastle University in England posted pictures of a pair of male eyes and the caption, “Cycle Thieves: We Are Watching You.” Bike thefts decreased by 62 percent in those locations — and not elsewhere.

One hypothesis is that officers got used to the cameras and became desensitized to them. But the researchers saw no difference in behavior during the initial phase, when the cameras were new. (The researchers also checked the data to make sure officers were turning their cameras on when they were supposed to, and found a very high level of compliance.) Another possibility is that officers without cameras were acting like officers with cameras, simply because they knew other officers had the devices.

An equally plausible explanation has to do with fear: In Washington, police officers are instructed to turn on their cameras whenever they answer a call or encounter the public in a law-enforcement context. The kinds of situations that might lead to civilian complaints or use-of-force incidents are high-stress encounters. When frightened, humans tend to act on automatic fear responses (or, in the case of good police officers in an ideal world, training).

“It’s a lot to ask, psychologically speaking, to not only remember the camera is on but to moderate your behavior,” said Yokum, the head of the Lab @ DC.

Finally, cameras may have had less impact in Washington because the police department there has already had to confront excessive-force problems. After a devastating 1998 Washington Post series revealed that the city’s police department had shot and killed more people per resident in the 1990s than any other police force in a large U.S. city, the Department of Justice entered into a memorandum of agreement with Washington to reform its policing.

“We went through a transformation with regard to use of force when Justice came in here,” Newsham said.

Cities that lack such accountability in their police culture may find cameras more effective, under this theory. (The Rialto Police Department had been reeling from a series of scandals when the Rialto study showed a large impact from cameras.)

Even if cameras do not reduce violent encounters, they can still offer other kinds of benefits: for training, or to hold a rogue officer accountable after the fact.

To Newsham, the cameras’ primary benefit is to improve relations with the community. “The transparency and trust that the community has, knowing your department is recording the interactions, I don’t think you can undervalue that,” he said. So far, it’s hard to say for sure if cameras increase trust, but Newsham said he’d like to find out more through additional studies like this one.

The nine-person research team pre-published its design online, so that there would be less temptation to rejigger the approach after the results were in (an emerging best practice in social science research).

“We like to be very anecdotal in policing — to compare this year to that year,” said MPD Commander Ralph Ennis, who oversaw the rollout of the cameras and worked closely with researchers. “This study is a whole other thing.”

© 2017 The New York Times Company