Three years after Ferguson, Missouri, erupted over racism in policing, we still don’t know exactly what led to the shooting that touched off the conflagration. We have video of the moments after a white police officer shot a young black man, Michael Brown. The footage shows Brown lying lifeless in the street, the horrified onlookers staring. But we have no record of the moments before.

In the absence of video evidence, two parallel and irreconcilable narratives exist. In one, Officer Darren Wilson was being threatened by a teenager who had recently robbed a convenience store, and the officer responded with lethal force when the teen reached for the officer’s gun. In another, Wilson shot an unarmed black teenager who may have offered some petty verbal challenge to his authority but posed no threat.

It’s possible that even Wilson doesn’t really know what actually happened; the literature on eyewitness testimony shows that it is appallingly unreliable. We think of memory as operating like a video camera, but in fact, it’s more like writing a short story: We edit, erase, reinterpret and confabulate — inserting details that support the narrative our brain has chosen.

The human brain isn’t likely to improve much anytime soon, but technology has. Now it is possible to have continuous video recordings from wearable cameras, so that the account of a moment like the shooting of Michael Brown would not be left only to memory. As America struggled to reconcile the two narratives about the Michael Brown shooting, many people suggested that mandatory body cameras on police officers would reconcile the starkly different views that white and black Americans have about law enforcement.

This made a lot of sense. We’d be able to prosecute any obvious abuses. Officers who used deadly force justifiably would be vindicated, no longer smeared by default as racist murderers. And the greatest hope was that if officers knew that they were being recorded, they would be more careful about their use of force.

It’s a hard goal to argue with. Even if you think that police abuse is rare, you will probably concede that a few bad apples are bound to get into any large barrel. But presumably, they are not going to show their rotten side if they know they are being watched.

Privacy advocates protested that police officers wearing cameras would be generating millions of hours of surveillance data with little transparency over how it might be used. Others asked whether certain kinds of police work might become more difficult. Who would want to offer an anonymous tip, or identify a local thug for a murder, if they knew that they were being recorded?

But the urgency of the questions about police violence outweighed these concerns, and by 2016, 95 percent of large police departments reported that they had either implemented body cameras or were planning to. My own city, Washington, not only moved forward with plans to put body cameras on its officers, but also set up the rollout as a randomized controlled trial, so that researchers could see what effects they had on things like use of force and civilian complaints.

That data is now in, and it shows that the cameras did … basically nothing. In fact, officers with the cameras had slightly more reports of use of force and civilian complaints than officers without them, though neither difference was statistically significant.

This is flabbergasting.

To be sure, I do not know how often officers overstep their powers, but given that they are human, I have to assume it happens sometimes. I also have to assume it would happen less often when they know there is a record that will call them to account.

This is just one study. But it is a big study, and has a very good design, much better than many earlier studies that purported to show benefits from adopting cameras. I would have expected to see some decline in measures like civilian complaints, even if small. I would certainly never have predicted that the cameras would not meaningfully alter anyone’s behavior.

The authors of the study offer a variety of explanations. Maybe even the cops without cameras behaved differently, knowing their colleagues could be recording. Maybe in the moment, situational factors overwhelm any worries about being observed. Maybe policing reform, implemented after earlier scandals about police use of force, has made Washington’s finest so very fine that they make not a single error that could be corrected by knowing that there’s an eye on them.

I don’t find any of these explanations particularly convincing, and I suspect the authors don’t either.

When we can’t adequately explain the data that contradict our deepest intuitions, another popular tack is to deny it. When the Oregon Medicaid study came out, showing that giving thousands of people access to Medicaid produced no significant improvement in measurable outcomes like hypertension control and cholesterol, many people immediately set about finding reasons that this study was no good, and should be ignored in favor of other studies (produced by comparatively fragile statistical analyses of observational data). This even though the only other randomized controlled trial on health insurance, a Rand study from 1982, led to a similar conclusion: free health care made people financially better off, but didn’t necessarily make them healthier. This result just seemed too incredible to believe, so people didn’t.

But the best tack when faced with challenging data is to open our minds, tell our intuitions to quiet down, and decide that we’re going to live in mystery for a while.

History is full of things that “we know, that ain’t so.” To rectify those errors, we needed to seriously accept the fact that our deepest intuitions, the things that seem so blindingly obvious to us that they’re not even really under discussion, may nonetheless be quite wrong — or at the very least, radically incomplete.

I don’t know why the Washington study failed to find any benefit from body cameras. Maybe it’s just random chance. (All a 95 percent significance level really tells you is that 1 in 20 times, even a sound methodology and careful statistical analysis will deliver you a false result.) Maybe some factor I don’t understand is skewing the data. And maybe, just maybe, those expensive, invasive body cameras really don’t make policing any better.

In which case the question is why they don’t. That’s where the above-mentioned explanations are useful: They give us some hypotheses to test as we start investigating our mystery. Eventually we’ll emerge from the other end of that investigation with a richer picture of policing, and human nature. In the meantime, we’re in for an uncomfortable wait.

McArdle is a Bloomberg View columnist.