« Scope of injunction in the 4th Circuit travel ban decision | Main | A Benefit of Magnanimity »

Friday, May 26, 2017

“Bright Data”  

At the core of the rise of data-driven policing is the ability to predict risk.  Predictive policing does not actually predict crime, but instead provides a mathematical assessment of heightened risk at certain places or with certain people.  The technology analyzes risk, and the inputs are generally identifiable factors that can be replicated across jurisdictions.

For example, certain environmental factors encourage crime in certain places.  These risks may involve fixed structures (abandoned property, empty lots), poor lighting (to avoid detection), escape routes (to avoid capture), access to legitimate businesses (to hide one’s intention), etc., with the growing number of risk factors adding up to a heighten likelihood of criminal activity.  Crunch the numbers, study crime patterns, and you can forecast where crime is most likely to occur.   

Similarly, certain socio-economic factors and lifestyle choices can heighten risk.  If you are involved in gang activities, drug activities, and live in poor areas in certain cities, your likelihood of being involved in a shooting is significantly greater than others without those risk factors.  Add in proximity to past violence, arrests, and past acts of gun violence, and particular people can be targeted as more likely to continue along that path.

At a very simple level (hidden by a lot of complex math) real world inputs get fed into an algorithm to create predicted levels of heightened risk.

In the policing context, this can be helpful to guide police to patrol areas predicted to be of higher risk, or to target individuals identified to be at a higher risk of crime. 

But, notice that the prediction of risk does not determine the remedy of a policing response.  All of the fancy risk assessment only goes to identify the places or people who might be at risk of being involved in crime.  The technology does not speak to how one might remedy that risk.  More after the break.

However, because the history of predictive policing has involved police (and been funded by police), we tend to think that the risk and remedy are connected, but, in truth, they are not.  One could just as easily send in an emergency urban planner into those hot spots of crime (with a grant to fix up the area, rebuild, add lights, provide economic opportunity).  Similarly, social workers, violence interrupters, and other community members could be sent instead of police to interact with individuals predicted to be involved in the next shooting.   Police do not have to have any role.  As I sometime joke when asked about my feelings about predictive analytics, “predictive policing is great, but probably better without the policing part.” 

One of my goals in writing about big data policing is to get people to see that the value in predicting risk can be acknowledged without necessarily also adopting a policing remedy.  In fact, it might be helpful to think about how big data risk identification would work without the police controlling the technology. 

I call this use of big data technologies to focus on environmental and social needs “bright data.”  “Bright” as in smart, revealing, or illuminating.  The idea is to turn the predictive analytics being developed to identify patterns of crime and look at trying to solve the underlying patterns of social risk.  We could – and some people are already doing it – map the social needs of society in the same way as we map crime patterns.  We could predict those in need of social services, as opposed in need of social control.

And, to be fair, some predictive policing models explicitly incorporate this social risk prediction as part of their overall crime strategy.  Under the leadership of Mitch Landrieu, New Orleans created a violence reduction plan (NOLA for Life) that combined big data technologies and predictive policing with social risk assessments and social programs.  But these big projects cost big money, and after initial success the reduction in violence has been difficult to sustain. 

I discuss this idea of bright data more in my book, and hope that the interest in big data policing will allow us to see the value of the technology in a broader light. At a minimum, I want the national conversation about big data predictive analytics to recognize that risk and remedy can be decoupled and studied separately. 

Posted by Andrew Guthrie Ferguson on May 26, 2017 at 11:49 AM | Permalink


Can police models predict whether or not crime would decrease if a country outlawed excise taxes, and decriminalized marijuana and vicodin? The answer may not only be more police with bigger guns, but also less laws or shorter sentences . . .

Posted by: skinnyPoliceModels | May 26, 2017 5:37:05 PM

Post a comment