There has been a media uproar over a new paper by Harvard economist Roland Fryer. In the study, Fryer looked at a number of data sets to assess racial bias in police use of violence. That's a natural topic to study, given the continuing trickle of videos of black people killed by police, and the backlash it seems to have produced, based on the lethal attacks on police in Dallas and Baton Rouge, La. The total number of Americans killed by police isn't in any official database, but the organization Mapping Police Violence has come up with some startlingly high numbers.

In general, Fryer found that police around the country are more likely to use many forms of nonlethal violence against blacks, even after controlling for all available factors, such as the behavior of the suspect.

But one of the Harvard professor's results didn't seem to fit this overall picture, and garnered an instant flood of attention. Fryer found that in Houston, black people involved in high-risk interactions with police — essentially, fights and chases — were no more likely to be killed than white people who were involved in similar situations.

This doesn't mean, of course, that Houston cops are less likely to shoot blacks than whites. Since whites are far less likely to be involved in these sorts of situations involving cops, their overall risk of death in confrontations with cops is much lower. What it means is that if you're black, and you get in a fight or chase with officers from the Houston Police Department, your risk of being shot is about the same as if you're white.

Does this mean that Houston police have no racial bias in their use of lethal force? Maybe. As Barnard economist Rajiv Sethi and MTV writer Ezekiel Kweku point out, if cops have racial bias, they might initiate more tense situations with black suspects who don't deserve it. Equal odds of being shot in an altercation might sound fair at first glance, but if black suspects who present no real threat to cops have the same chance of being shot as white suspects who are really threatening the police, that's not fair at all. Sethi and Kweku are raising the possibility that Fryer's result suffers from selection bias.

Fryer, being a top economist, is used to grappling with this sort of issue. He's well aware of the possibility of selection bias, and attempted to address it in his paper. He tried looking only at situations in which a robbery or violent crime was in progress, and found the same result — race didn't affect the odds of being shot by Houston cops once an altercation started. Of course, that leaves the possibility that the result isn't relevant to things like routine traffic stops or street searches — the kind of encounters responsible for the killings seen in recent videos.

So what lesson should we take away from the Fryer study? We shouldn't be too quick to extrapolate the Houston result to the whole nation. There's probably a huge amount of variation in how biased different police departments are. But we should also avoid the temptation to dismiss the result out of hand, as some writers have done. If Fryer's Houston finding — which he calls the most surprising of his career — tells us anything, it says that to reduce police killings of black people, reform should focus not on violent standoffs, chases and fights, but on routine interactions involving traffic stops and street encounters.

This is the low-hanging fruit of police reform. There are at least three obvious things that can be done. First, racial profiling in traffic stops needs to be reduced. Studies in Vermont, California, Illinois and other places around the country have documented racial bias in traffic stops and in stop-and-frisk encounters. Training police to recognize their implicit tendency to be more suspicious of black motorists and pedestrians would cut down on the number of opportunities for violence against black people.

The second is to reduce police use of lethal force in general. Although crime has fallen a lot in the U.S. since the early 1990s, most measures of police shootings have stayed roughly constant. There should be a comprehensive nationwide effort to discourage police from shooting people, and to aggressively prosecute officers who shoot unthreatening and unarmed suspects.

Finally, a national review should be conducted to identify which police departments suffer from racial bias in their use of lethal force. Better and more uniform data, collected by the federal government, would allow Fryer's Houston analysis to be replicated elsewhere, and extended to situations like traffic stops. Fryer's national-level data shows that racial bias is common nationwide, but finding where it's concentrated would allow quicker and more effective action to correct it.

If the U.S. takes these steps, it would go a long way toward reducing the dismaying problem of police killing of black people, and quiet much of the unrest that has exploded in the last few years. If that happened, it might even help police feel safer doing a difficult and dangerous job.

Noah Smith is an assistant professor of finance at Stony Brook University and a freelance writer for finance and business publications.