You have cops everywhere doing cop things. Every now and then a cop does something really bad and someone gets killed. That is not going to show up as a statistical effect when looking at the overall behavior of the cops, unless you do some very specialized Bayesian statistics, which no one is going to do on that sort of data.

So, the just released study, summarized here by NPR, is not a surprise.

Having police officers wear little cameras seems to have no discernible impact on citizen complaints or officers’ use of force, at least in the nation’s capital.

That’s the conclusion of a study performed as Washington, D.C., rolled out its huge camera program. The city has one of the largest forces in the country, with some 2,600 officers now wearing cameras on their collars or shirts.

“We found essentially that we could not detect any statistically significant effect of the body-worn cameras,” says Anita Ravishankar, a researcher with the Metropolitan Police Department and a group in the city government called the Lab @ DC.

“I think we’re surprised by the result. I think a lot of people were suggesting that the body-worn cameras would change behavior,” says Chief of Police Peter Newsham. “There was no indication that the cameras changed behavior at all.”

Perhaps, he says, that’s because his officers “were doing the right thing in the first place.”

The purpose of the body cam is to find out what happened in this or that extreme event, when there is a death, or a plausible accusation of wrong doing. Another likely effect of the body cam is to cause bad behavior to be more rare, just because the cameras are there. But bad behavior is often the result of out of control emotions, or of bad training, or some other thing that the presence of a cam is unlikely to detect. I would not be surprised if body cams did have this effect on some police forces, but not most.

There are statistical ways to see if the body cams are having an effect on behavior but they have to assume that the behavior is a) really there already and b) of a certain specific form. Then you can subset your data down to the sensitive parts of it, and see if, say, before or after body cams you get a difference. But really, the best way to do that is to pair up the cops (statistically) so you have one set of cops match, with respect to relevant characteristics, another equal size set (on the same police force) where one set is wearing the cams, the other not (there are some added difficulties in doing this, like matching patterns so neither or both have the cams, etc). This sort of approach could identify an effect. But the methods that would normally be used won’t unless it is dramatic.