The national protests against the police misconduct that led to the death of George Floyd in Minneapolis have generated fresh accusations of misconduct by officers. Phone cameras have captured police firing nonlethal bullets at reporters, firing paint canisters at people on their porches and driving recklessly into protesters.

Minneapolis police officer Derek Chauvin has been charged with third-degree murder and second-degree manslaughter, and three officers who were at the scene and did not intervene were fired (but not charged). The uncountable number of Americans who have seen the video of that incident — and who afterward saw police unleashing violence against protesters, including in front of the White House — may despair that police reform can ever work. Indeed, much more academic work has been done in estimating the value that policing brings to society — a review by Rand Corp. finds hiring an additional officer saves an average of $437,000 through crime reduction — than in measuring the costs of unfair policing on communities of color.

The most important reforms, of course, will arise from listening to the people who suffer from unjust police practices. But some academic work points the way to potentially promising changes to how police do their work that could lead to better outcomes. In 2018, for example, I and three colleagues published a study finding that scheduling brief meetings between officers and sergeants, where they talked through encounters on the street, led to more measured responses to later incidents. The conversations were explicitly designed to make officers more conscious of the importance of “procedural justice” — that they should listen to each disputant, be open to correcting their own errors and keep an eye out for turning points that can escalate or de-escalate a situation.

When we examined the subsequent actions of those officers, those who met with their supervisors were 12 percent to 25 percent less likely to resolve incidents with an arrest than members of a control group, and around 40 percent less likely to be involved in incidents where force was used (very rare events to begin with).

A total of 180 officers were selected to take part in 221 meetings, which took place over six months in 2013, at a time when the Seattle Police Department was being monitored by the U.S. Justice Department because of a pattern of excessive force against nonwhite citizens. After each meeting, we compared the officers’ performance in the field with that of a set of otherwise identical control officers.

One way we thought the program might work was by “slowing down” officers’ thinking on the streets. In general, the more experienced people are, the faster they work and the fewer mistakes they make, relative to newbies. This is because once you do a task many times, you don’t need to pay attention to each specific thing; your actions become automated. But in police work, such “efficiency” can come at a cost: Officers may get so used to handling routine disputes and low-level violations (loitering, jaywalking) that they stop needing to listen carefully, and this may make adapting to distinctive scenarios more difficult. (George Floyd’s arrest began with a dispute over a possibly counterfeit $20 bill.) In contrast, to do procedural justice policing right, citizens need the opportunity to explain themselves and correct mistaken impressions the officer might have. This requires constant attention and information-gathering on the part of an officer — the opposite of mental automation.

Our hypothesis was that the very thing that helps experienced officers do their jobs well made it harder for them to catch the mistakes procedural justice is focused on correcting. The light-touch intervention we designed is aimed at pushing back against some of the automation that comes naturally to good officers.

Police work is often defined by numbers and hierarchy: Officers are evaluated based on crime levels, arrests and “clearance rates,” and they take orders from their superiors. The conversations we designed with the department’s deputy chief were intended to redirect officers’ attention to how they were interacting with both perpetrators and witnesses. Sergeants asked open-ended questions about how the officers learned about the call; what they saw when they got there; how they used that information to make decisions, and what they might have liked to change, if anything, in how the dispute was resolved. As part of each discussion, sergeants asked the officers to evaluate the sergeants’ own role in the conversations, as a way of modeling openness to feedback.

Each conversation took less than 20 minutes. These meetings did not supplant the department’s disciplinary process: No incidents subject to traditional oversight — as when an officer used force or a complaint had been filed against him or her — could be discussed. In anonymous surveys, officers said they tended to like the meeting, even if they sometimes wished they could be talking about more exciting incidents.

We evaluated the outcomes of the meetings using administrative data already collected by the Seattle Police Department. First, we found no evidence that this extra meeting led to “depolicing.” It’s sometimes thought officers put under special scrutiny will respond by doing less, to avoid getting into trouble. But officers in the program responded to the same number of calls and were equally likely to engage the public than officers in the control group. There was no evidence officers in the “treatment” group spent less time on scene or wrote fewer reports. There was also no evidence crime went up.

Officers who had these conversations, however, made different decisions in the field. When we compared officers’ officially recorded activity one and six weeks before and after the meetings — and explored how the control group fared over the same period — officers who were reminded to “think slowly” were significantly less likely to make an arrest and less likely to be involved in an incident where force was used.

This is far from the final word on the subject. For example, we didn’t end up with enough data to draw any conclusions about whether civilian complaints went down. Limitations in the data Seattle collected at the time also meant we could not evaluate how officers engaged with people with different racial and ethnic backgrounds — a crucial element in Floyd’s death and in the subsequent protests. As with all experiments, this one needs to be replicated in many contexts before we can draw firm policy conclusions. Caution is especially warranted when studying rare events like use of force and civilian complaints.

Still, the findings offer some hope that a relatively minor intervention can potentially make a real difference in the quality of policing. That may seem hard to believe now, with all the chaos on the streets, but when the clashes die down the hard work of reform will begin anew. If the results of this experiment hold in other places, we can have public safety with fewer arrests — and less violence.