A little while back, I wrote something on the research of Andrew Papachristos, a Yale sociologist not long out of Chicago, on the small social networks of Chicago homicide. In short, how much of fatal violence in Chicago is contained within a comparatively small group of people, who are themselves linked by crime.

It was a fascinating paper, but it had its limitations. Chicago has high raw numbers of homicides and a relatively high homicide rate. But statistically speaking, homicides are a small sample of violence in the city, and in some respects a random one. Recently, Papachristos followed up with a new investigation, and a much larger new data set: non-fatal gunshot injuries in the city.

The data in the new paper is equally fascinating, and on one level, as you might expect, quite troubling. To begin with, the dramatic disparities the rates of nonfatal gunshot injury: overall it's 46.5 per 100,000 for the city as a whole from 2006-2012. It's 1.62 per 100,000 for whites; 28.72 for Hispanics, and 112.83 for blacks.

For all males, it's 44.68 per 100,000; 239.77 for black males, and for black males from 18-34 it's 599.65. As Papachristos and co-authors Christopher Wildeman and Elizabeth Roberto point out, that's a staggering one in 200.

The numbers are enormous, and they've caused a lot of pessimism. But the point of digging into the data is to create, literally, maps—to follow the violence through the city and, as maps are meant to do, guide us back to its sources. And in that sense, cohesive patterns emerge.

Papachristos constructs a social network—not a virtual one in the Facebook sense, but a real one of social connections between people—by looking at arrestees who have been arrested together. That turns out to be a lot of people in raw numbers, almost 170,000 people with a "co-offending tie" to one another, with an average age of 25.7 years, 78.6 percent male and 69.5 percent black. It's also a large percentage of all the individuals arrested: 40 percent of all the individuals arrested during that period.

Within the entire group, the largest component of that whole co-offender group has 107,740 people.

Within the timeframe—from 2006 to 2010—70 percent of all shootings in Chicago, or about 7,500 out of over 10,000, are contained within all the co-offending networks. And 89 percent of those shootings are within the largest component.

Or, to put another way: the rate of gunshot victimization (nonfatal + fatal) in Chicago is 62.1 per 100k. Within a co-offending network, it's 740.5—more than 10 times higher.

"This finding has (at least) two implications for our understanding of non-fatal gunshot injuries," the authors write. "First, the concentration of non-fatal gunshot injuries in networks such as these demonstrate that such incidents are more concentrated than previously thought, and even more concentrated than gun homicide by either demographic group or place. Our findings indicate that 70 percent of all non-fatal shootings occur in networks comprising less than 6 percent of Chicago's total population. This distribution of shootings within co-offending networks fundamentally changes how we assess the distribution of risk in Chicago." (Emphasis theirs.)

That, from the perspective of a sociologist or criminologist, is reason to be optimistic, in a sense. Density, and density of connections—especially given the comparatively blunt data that is arrest records—means paths, and paths can be followed and refined.

Papachristos spoke with me about his research, its implications, and what happens next.

Compared to your last paper on the social networks of homicide, scaling up to nonfatal gunshot injuries involves much more data. Did anything surprise you about the findings?

The thing that happily surprised me—from the statistical standpoint, not the practical standpoint—you start to see more of the variations you would expect to see by gang membership and race. One of the things that had me a little anxious in the first paper was that the gang member variable wasn't significant. And I hypothesized that it was because there was so much uniformity in the data in the West Side.

In the AJPH study, everyone was black, 40 percent were gang members, and they were all young men; they were all exposed to the same neighborhood conditions as well. When you expand to the whole city, you get Rogers Park, you get Englewood, you get Little Village. You get different types of conditions—housing projects, no housing projects. And that puts a lot more variation in people, and they're exposed to different things. If everybody's coming from the West Side, there's still some variation, but by and large they have the same institutions and sorts of structures.

What you see when you expand to the rest of the city, into the nonfatal shootings, you start to see that work like you'd expect it to work.

The thing that really shocked me was how connected the city was as a whole. The largest component, the big network in the city, is like 106,000 people. That includes people of different races, from different parts of the community. (The study I'm working on now actually looks not at the connections between individuals, but between neighborhoods.)

It means that the city is connected in crime in a much more complicated way than just being spatially next to each other. I anticipated that you would have a bunch of small networks—this neighborhood, this park, South Side, North Side, all the things we usually think about. But what you see is that there are these connections that bind this world together. But it's not some kind of organized crime syndicate. We're talking about everyday interactions—small stuff that links people together. I never thought you'd have such a massive chunk of the city connected to each other.

I was interested to see what would happen when you move from homicides to nonfatal shootings, you're kind of moving to a less random outcome.

Not just less random, but more of it. From a statistical standpoint, you have fewer zeroes. One of the things that happens is that it's easier to model. It's like when historians or political scientists model wars, but there's a small number of wars and conflicts, statistically speaking. The same is true of homicides. Even though there are too many homicides, they're about six to one fatal to non-fatal. Not only are they less random, they provide you with the statistical power to do things you couldn't do with fewer cases.

We all know this, but because of how we track data, it's easier to track a homicide than a non-fatal shooting, just from the way things are reported.

That's always surprised me a bit—the things that police would want to track, and public health agencies would want to track, the data's just not as good.

It appalls me, actually. I wish you could get data on nonfatals. It has a lot to do with resource allocation and tracking. If you have to spend investigatory power on a homicide versus a shooting, it usually goes to the homicide, for all sorts of political, but also practical reasons.

The other thing is the legal classification—you guys ran that big story about all this. The classification of shootings is even more tricky than murder, homicide, manslaughter, and all that. It can be an assault, an accident, all sorts of things. In a lot of cases, what happens where they don't have an offender, it's not quite clear—and this is my understanding from the outside—it's much less clear how you classify it.

Victim records, when the individual is not dead, aren't necessarily public, unless they waive their rights. It's not the same as a death certificate, which is a public record. I don't want them to die, but it's a practical barrier. The victim doesn't have to give their name, necessarily.

When I read this paper, I felt a lot more pessimistic than the previous one, because the numbers are so much larger—not just in terms of victims, but also how the odds increase when you scale up. My reaction was that it makes the problem seem more intractable.

I didn't have that same reaction. The one thing I found most reassuring as a scientist is that the basic, same finding was there. It's still super-concentrated within small networks. It suggests, again, short-term remedies—not long-term remedies—being more directed and more precise.

The problem that might make it seem intractable is that there are just so many more. But the same central premise is true, which I think is a good sign. In addition to digging even further, and trying to find more about the pairing of victims, that's obviously the next step. The good news is that it's not the majority of people—less than eight percent of the population in these communities.

This focus on only homicides is a bit off. There is some statistical randomness—we would be better off focusing on non-fatal and fatal shootings, because there aren't that many differences between them, other than a couple inches. And we don't really know a lot about the survivors. That's one thing I don't go into, except at the beginning of the paper—we don't know what happens to these people who are living with bullets in their bodies, they have all sorts of stress. It's been getting a lot of press these days, and I'm glad it is—PTSD, anxiety, disabilities.

We really don't think about that. If we want people and communities and families to be healthy, we need to pay attention to that in a different way. That's something where I hope the media as well can start to do, and I think they have, especially on this issue of exposure to violence.

It seems like there's progress on that, such as Jens Ludwig's work on cost-benefit analysis with educational programs, but the data is really broad.

There's no intersection between the different levels of exposure [to violence]. Even in my own work, the stuff I've done with networks, the stuff I've done with Patrick Sharkey where we looked at kids in the Head Start program who'd had a homicide on their block, we know it does bad things to kids, we know it does bad things to kids, we know it does to your brain if something happens in your neighborhood, we know if you hang around people you are more likely to be shot, we know that exposure at community and family level is related to all sorts of bad things.

But exposure to different levels of violence, in different ways, we haven't begun to untangle that. It's hard to think at which level interactions should come in. Jens's stuff is focused on this really vulnerable age, adolescence. If they stay in school, their trajectory is quite different, even without such programs, but we want to funnel resources into those individuals.

We also have to think about the systemic community level—how do we improve communities? More at that macro level. Healthier, safer communities. All these things at once; it makes it so hard.

That's what I was thinking when I was reading our piece on crime statistics—whenever there's a spike in homicides, all the attention goes to the cops, and it's a narrow way of looking at the problem.

This is a great example of how we think about health care and health. This last series of mass shooters—the discussion went to mental health, for good reason. We don't want crazy people with guns; that discussion should be had. But the real reason we should be talking about mental health is, what does the mental health of, say, mothers does for their families. Even in adverse conditions, even in high-crime communities, high-poverty communities, disadvantaged communities, if the mother of the child is mentally stable and healthy, the child does better. We know this. We know this from clinical trials. We know this from all sorts of things. Yet we never discuss that as part of safety and childhood well-being.

I think part of it is, and this is not to blame the police, and the public more generally, is that we want responsive policing, we want them to come when we call 911 or 311, and if they don't we're angry. And that the same time we want them out there building community, preventing crime, which is not really what policing is designed to do, though that's what they spend a lot of time doing.

But we don't really look at how health centers or schools can prevent crime. This is the one thing with school violence are about preventing crime tomorrow. Yes, they want to reduce violence today, but even the evaluation with [Becoming a Man], part of that was the kids having better educational outcomes. Not just about the decline in arrest rates, which was important—they're not the killers or shooters. It happens. But if you look at the average age of shooters and victims, it's in the 20s, not in the teens. They happen, but that's not what you're doing. You're preventing something later.

One last example: the Brady Bill, from the early '90s, everyone wanted that to be a gun-violence reduction program. And it actually did almost nothing for gun violence, but it did a whole hell of a lot for gun suicides. Some people say that's a failure; I say that's a great success because there are more suicides than homicides. It didn't have the intended effect, but it had very dramatic effect on another form of violence. So we need to expand the ways to do the safety and health of communities. It involves policing, but it also involves these other things.

It becomes difficult when you have competing interests and competing budgets, trying to figure out how you're going to frame something. The White House has put out a call for these types of mentoring programs, and that means people are going to shift how they frame their work and what they're going to be doing. That happened when we did the faith-based communities under the early Bush administration as well. That's one of the things we need to do is think more holistically.

It's also hard, because there's no one agency that does it all. Police and schools are the two go-tos, but there are things they just don't do. Trying to build some kind of collaborative front is hard to do.

What happened with the Brady Bill and suicides/homicides?

The Brady Bill essentially called for waiting periods for handguns. Three to five days, I forget exactly. But that was the central feature. As you can imagine it started this entire debate—is that violating our civil rights, we won't be able to protect ourselves. It was ruled constitutional, and it came into effect. So from then on you had to have this waiting period to buy a gun.

The subsequent evaluation of the Brady Bill, some people say it works, some people say it doesn't, or if it does work, it has very modest effects on aggregate levels of gun crime.

What people consistently find is that there were pretty significant effects on gun suicides—a decline in gun suicides since the Brady Bill. And the logic is something that people who study suicides know: when people kill themselves with a gun, they don't spend a lot of time thinking about it. They go through a horrific mental process, but when they actually get the gun and pull the trigger, it's actually minutes, it's not weeks or hours. So it makes sense that if you have to wait to get a gun, it would have an effect on gun suicides.

And gun suicide rates had been unlike gun homicide rates, which kind of go up and down with trends. Gun suicides have stable patterns. They tend not to run in the same direction, ups and downs, of homicides. So it was a startling change. Unlike the effects of gun crime, which mainly we're talking about homicides, the effects and studies on gun suicides and the Brady Bill were pretty consistent.

One thing that comes up in the paper is that there's a need to better understand the "contours" of social networks. What does that mean, and how do we get there?

It's like looking at a street system. Networks can be built a bunch of different ways. It can be a grid like Chicago, it can be whatever Boston is, I can't even describe it—spoke-y, hub-y, crossing paths, backroads. They look different as a network.

Part of what I mean by understanding the contours is figuring out if this whole thing is really a super, big clumpy plate of spaghetti, or whether it's spread out. Whether or not there are dense pockets, and that those might be gangs or not gangs, or loose kinds of little triangles.

There are also statistical properties of networks. Is it a small world? This is one of the things I'm working on with a student now. Small world graphs—this idea that creates that phenomenon where you say "oh, you know so-and-so? What a small world." There are statistical properties that include clustering and path length, literally how many of these handshakes away you are.

So when I say "understand the contours of these graphs," what I actually mean is "can we do a better job of mapping them out?" Can we say, this is what the criminal world actually looks like? All I've done at this point is create the damn map. Now it's time to figure out, are there certain structures, for example, certain road patterns, that are more conducive to violence than others?

To draw the analogy with suicides, for example, there have been a few studies on networks and suicides among schoolkids, for example. Obviously one of the things we know—we don't know, but intuitively, but it bears out that people that are in very isolated networks, without a lot of ties, have a much more increased chance of suicide.

So one of the things you think about in terms of how that relates to prevention of suicide is making sure people have ties, whether that's from the school, or fostering friends, or encouraging whatever, but that actually plays out in suicides.

We don't know the same just yet for the gun violence issue. Are there certain types of paths, or network structures, that are more conducive to victimization? There are all kinds of hypotheses. One of them is being in between. For example, if you're in between different parts of a network, if you're that person who connects a network, perhaps you're more exposed. This is what ethnographers or cops tell you about gangs. The guys that go between factions, or running with two different crews, are a little more at risk.

That's what I mean by the contours: trying to understand more precise ways the network could effect these risks.

How do you think people will go about finding these contours?

First of all, I hope they do, let me start with that.

Now, I'm a smart guy, but I'm not that smart, though. All I've really done is take one developing mode of science and apply it to that area. What people need to do is something quite similar—we know a lot about how networks effect all kinds of health behaviors; suicides, like I mentioned, but also depression, happiness, disease transmission. We just need to do the same steps at this point.

I've laid out in this paper ways to do these networks, it just some more people to kind of go at it. Because this is a really wide-open area. The more people that start to do that, and analyze those things, the more you'll find. We have plenty of places to start.

The danger, of course, is realizing that we're talking about violence and arrests, not Facebook and the flu. You need to remember what these networks are. That's what happens when I give presentations or talks—everyone  assumes I'm talking about Facebook or Twitter. And I have to remind them that I'm talking about behavioral networks, people doing stuff.

Among the people in these networks, how good a sense do you think they have of the risks? I'm assuming that if you're associated with a lot of people who are committing crime, you do, but if you're on the margins, do you realize how much the risk is increased?

The statistical and practical experiences, for me, have been quite convergent. When I show these to people, young men who are involved in these interventions and the strategies we're working on, they're riveted. And they're riveted because of the same things that you and I do when we're talking about our networks.

They know their friends—"oh, yeah, I know these guys." And then they'll forget someone—"I forgot about that guy." Like the movie Hangover: "I forgot about that, we got arrested, it was crazy." More often than not, they remember their immediate connections. The issue with the world is not that you don't remember your friends and associates, the issue is that you don't always pay attention to their associates.

That's why this exposure, and those distance measures, are so significant. Again, this goes back to other sorts of phenomena, like the small-world phenomenon. When you have that feeling—"oh, you're so-and-so's cousin," or "you know Tommy," or "I forgot you were at that party, because that was so-and-so's party." That's when you start to realize that people don't see that far in their own networks. It's true in all kinds of networks. I know who I'm sleeping with, and my partner might tell me who they were sleeping with, but I probably didn't meet them. You don't actually see very far. But just because you don't see it, it doesn't mean it doesn't effect you, and that's actually what puts people at risk.

People have a sense of these networks. As an example, I'm working on this New Haven project—there were three homicides that just happened, young men, 16 and 17 year-olds, these last two and a half weeks. All three of the victims were connected. Not only were they connected to each other, they were connected to two victims last year. So they all knew each other. They were not necessarily related instances; they weren't retaliations or some Romeo-and-Juliet thing. But this is their world. If you asked one if he knew the other, they'd say, yeah; they went to the same school.

People really do know who's around them. The issue is not that; the issue is when your friend says, somebody came and shot at me. You don't know that somebody, but because that person's your friend, you get pulled into it. That's true in all sorts of behaviors, but particularly in this instance.

When they see it, they become riveted. You become obsessed with it—or maybe I do, because I study it.