Predicting Crime: Is Stealing a Car Like Choosing a Restaurant?

 

Dr. P. Jeffrey Brantingham is a professor of Anthropology at UCLA and an expert on crime analytics. He also serves as Director of University of California Mathematical and Simulation Modeling of Crime project (UC MaSC), and as Chief of Research and Development at PredPol, a crime analytics consulting firm. His 2012 paper, “The Ecology of Gang Territorial Boundaries,” was covered previously by the Chicago Policy Review.

 How predictable is crime?

It’s much more predictable than people think it is. Victims of crime often feel like the crime has come out of the blue. If you’re robbed at gunpoint on the street, it feels like a random event. It turns out that crime has patterns: collections of where and when crimes occur, and maybe some of the whys, are quite regular. How predictable crime is varies from crime type to crime type and location to location. There always is a fair amount of noise in crime data, but the prediction problem is quite tractable.

When you look at the ethnography of crime, there’s always an individual story that goes along with an individual offender or an individual victim. But when you take a step back from individual stories, there are a lot of things that occur in common. No matter where you go, Europe or Asia or North America, if you ask burglars, “why do you break into these houses?” they’ll give you lots of answers that point to a generalization: “I broke into that house because it was easy. And here’s why it was easy.”

They report the same sorts of reasons for committing the crimes where they do and when they do. As an anthropologist by training, I’d say there’s a really good reason for that: we are evolved to solve foraging problems, and many offenders are just applying those things that come naturally to people.

We can say a lot about how far people are willing to travel to commit crimes. It turns out to be not very far at all. Most offenders commit crimes very close to where they work, where they live, where their girlfriend or boyfriend lives, or the anchor point for their activity. That’s pretty much universal, and there are good ecological reasons for it. I would imagine there are policy implications for that, be it policing strategy and tactics or urban development and design. Those sorts of things are universal.

What types of crimes are most predictable and why?

Homicides are inherently less predictable, just because there are so few compared with something like car theft. So there’s a sample size issue. Some crimes also have much stronger connections with environmental characteristics. You can only steal cars where cars exist in undefended locations. So a shopping mall that every day has 5,000 vehicles in unprotected circumstances produces a signature that leads to pretty good predictability about where and when car theft is going to occur.

Homicides are also predictable. It’s not uncommon to have a domestic homicide preceded by four or five domestic disturbances. Police officers know this. In the UK, the law is such that if [police] come across a domestic disturbance situation and one of the parties makes a clear accusation, then the police are obligated to separate them for the night. These things follow very regular patterns of escalation that can be statistically documented.

How do you show that prevented crime isn’t being just pushed into other neighborhoods?

That is what’s called “displacement,” and it’s been a hot topic in criminality for decades. All of the experiments that have been done on displacement have pointed to a conclusion that is incomplete. For some offenders, if you take away their preferred location, they will desist for a time. In the time that they are desisting, you will get fewer crimes.

Now, there is a key distinction here: the difference between preventing crime and preventing criminality. If you take someone that has a serious drug dependency and they are committing property crimes for the purpose of supporting their drug habit, preventing them from committing a crime today is a benefit to the community because you have one less crime today. But in so doing, you haven’t solved the problem of that person’s drug dependency. That is a completely separate issue. And completely different issues are needed for that problem.

What we’ve been working on is about preventing crime, rather than criminality. Preventing criminality is potentially a much harder problem. Criminality is multi-causal—there are lots of reasons why someone is willing to commit crimes in the first place. One of the big challenges is that criminality is something that occurs over a lifetime. So with a lot of policy ideas, if they get put into place, you’re not going to know how they work for 8, 10, or 15 years. That’s a huge challenge.

There’s no easy solution to that problem. That’s not what we’re working on. We’re not about preventing criminality; we’re about disrupting opportunities for crime in the here and now.

What privacy and constitutional concerns do you think policy makers need to take into account as they apply crime prediction?

Policing and policy have to be developed within the constitutional framework and there will always be limits on what can and cannot be done within the context of law enforcement. A lot of the work we do on prediction is not about predicting who is going to commit a crime, but about where and when crime is likely to occur, regardless of whom the offenders are.

What it’s really focused on is disrupting opportunities for crime and preventing crime. Preventing crime benefits everybody. It benefits potential victims because they don’t come home and find their car stolen. And although they might not see it this way, it benefits potential offenders. If you’ve prevented them from committing a crime, that’s one less chance for them to run afoul of the legal system, and that does benefit them. It also benefits police because it reduces the amount of time they have to spend processing crime scenes or arresting people and engaged in costly procedures. Crime prevention benefits everybody. I can’t see how that wouldn’t be an appropriate direction to move.

What we’re doing is about predicting where and when crime is likely to occur. My understanding is that that information is usable within the context of current policing for police to engage in further investigation. Police always have to have other information and observations at their disposal to engage in police action with people. I think it fits squarely within the legal arguments for hot spot policing. A lot of what we do with predictive policing is just a much more refined approach to hot spot policing. In some ways, you might think of it as being a step in the right direction. It actually defines, in a much more limited and precise way, where law enforcement is needed.

This is not so much about moving police from one area of the city to another; it’s about saying in a given area, where you have a given amount of resources and a given amount of time, where you should be putting those resources.

In your paper, The Ecology of Gang Territorial Boundaries, you used a model of predator-prey interaction to explain gang violence. The model you are using in your consulting work with the LAPD is based on earthquake prediction models. How did you and your team decide to use models from the natural sciences?

I don’t find it surprising that models that you’d use to study the territoriality of birds, chimpanzees, or honeybees could be used to study humans. Evolution has solved the same sorts of problems over and over again and arrived at the same sorts of solutions over and over again. If there’s a really good solution that works for birds and chimpanzees and honeybees and many other hundreds of species, I don’t see why it should be any different for humans. There are going to be twists and turns for each species, but, having said that, there are many generalities that are consistent across all those groups.

We have a diverse team that’s familiar with lots of things. I’ve done a lot of work on ecology and was in a position to make those connections across those different domains. It requires being familiar with all those areas, finding the right sorts of connections, and adjusting the models to deal with the new situation. There could be other models that are just as good or better, but it is a process of discovery. You have to go out and look for it and do the careful science to put those models to the test.

What applications are there for these models beyond crime?

Lots. The public in general thinks of crime as a special category of human behavior; I see it as the exact opposite. It’s part of normal human behavior. How and why offenders select places to break into or cars to steal, follows rules that are very similar to how you and I choose which restaurant to go to. You can make direct comparisons across all that. Crime fits into normal human behavior. The implication is that there are all sorts of predictive analytics you can do.

Feature Photo: cc/(Alan Cleaver)