When Bill de Blasio first ran for New York City mayor four years ago, ending “stop-and-frisk” police searches was a cornerstone of his campaign. Critics warned halting the practice would fuel crime. But this week de Blasio coasted into reelection against a backdrop of historically low crime rates.

The city of more than 8.5 million people has seen fewer than 300 murders so far in 2017. That puts its body count lower than much-smaller jurisdictions including Baltimore, a city of fewer than 620,000 people where 303 people have been murdered this year, and Chicago, where the number has risen above 580 in a population of 2.7 million.

So what factors can really help drive down crime? The National Academies of Sciences, Engineering and Medicine said in a report released Thursday that certain “proactive” policies aimed at preventing crime before it happens—including stop and frisk—show mixed results. Yet it is not enough to simply identify what policies appear to reduce crime, a panel convened by the National Academies cautions in the report. Authorities must also consider the real-world risks of applying these approaches in ways that are racist, biased or illegal, they wrote.

Historically, policing has focused largely on responding to calls and investigating crimes. But in the past few decades there has been a shift toward preventing crime by routinely sending officers into communities and identifying potential problem areas. Not all police departments are using these strategies, notes David Weisburd, chair of the expert panel and director of the center for evidence-based crime policy at George Mason University. But it is becoming relatively common and is a big departure from the standard model, in which police mostly respond to crimes that already occurred, Weisburd says.

What Works

“For police chiefs who want to do something, increases in violent crime are often very localized and occur among specific people and on specific streets—and the evidence from the report is that when you focus on those, you can produce reductions in crime,” Weisburd says. “Hot-spotting,” for example—a practice in which police are disproportionately stationed in areas with higher crime rates—seems to help, and does not just displace crime into immediately surrounding areas, the committee says. And stop and frisk can be effective when it is highly focused on areas with high concentrations of crime or robberies, Weisburd adds. His committee also found that third-party policing—in which businesses or building owners partner with police or are pressured to work with them—can help. When police officers identify specific problems, try to understand them and make a tailored plan to solve them, it can reduce crime, too. Finally, focusing police resources on high-rate offenders (to either get them off the street or reduce crime) has good evidence behind it.

What Doesn’t Work

The report also identified police strategies that do not seem to work. “Broken windows” policing, in which officers crack down on even small instances of disorder before they overwhelm a neighborhood, does not typically lead to less crime, the panel says. But it adds that if such efforts are much more nuanced—focused on a small number of high-crime streets—they can sometimes make a positive difference.

Another topic under the panel’s scrutiny is community policing, which generally refers to police building relationships with local residents and involving them in their decision-making about problems. Politicians and others have pushed the concept hard, but the panel says it does not have strong evidence that community policing reduces crime. “If you can increase cooperation with the public, you would assume they report crime more often. But programs so far that encourage community policing—that use newsletters to involve the public, meet often with the public and spend a lot of time dealing with the public in a cooperative way—those projects, at least from the evidence we have now, don’t seem to have crime prevention effects,” Weisburd says. There is evidence these programs improve community attitudes toward police, however. “So when police are thinking about what they are doing, they should think about it as a method to improve relationships with the public and perhaps combine different strategies together to reduce crime and improve community perceptions,” Weisburd adds.

“I wouldn’t take [the panel’s report] as torpedo below the waterline for the entire concept of community policing. I would say we need a longer study of what types of community engagement strategies work,” says Ames Grawert, a counsel at the Brennan Center for Justice, who was not involved with the report. “Our hunch is that, over a very long time, strategies focused on getting officers engaged in the communities they serve will pay off in a number of ways. But I don’t expect to see significant changes in just a year or two.”

The report’s findings are somewhat at odds with some of the conclusions from The President’s Task Force on 21st-Century Policing, whose recommendations in 2015 included a strong push for procedural justice—which seeks to convince communities, through its interactions with them, that the police are neutral and exercise their authority in legitimate ways. Weisburd notes the task force promoted this approach, but his panel concluded there was not strong evidence to support it. “We found there isn’t enough evidence to say if it has an effect on crime, and whether or not to say it has a strong effect on perceptions of legitimacy,” he says. “We aren’t saying it doesn’t have an effect, but we are saying that there isn’t strong evidence.” With longer-term data, the Brennan Center’s Grawert says, there may be more positive results.

The Jury Is Still Out

Most studies on police policies have focused on short-term evaluations, so the committee could not point to what will work over longer periods, Weisburd explains. It is also hard to know how well interventions applied across an entire jurisdiction will work in particular neighborhoods or if strategies that work well in one area will definitely work well in another. There is also very little social science data on how often proactive police tactics lead to illegal behavior, power abuses and differences in outcomes by race. “Right now,” Weisburd says, “we just don’t have that evidence.