Nic Garcia
Dallas Morning News
DALLAS — A casual glance at a crime map of Dallas suggests the violence that barreled through the city last year was aimless.
But a more detailed look at North Dallas reveals a high-risk setting for gun violence stems in part from a concentration of used-goods stores, apartment complexes and car washes. In southeast Dallas, the constellation of motels, gas stations and pharmacies serves as another prime location.
Those conclusions were based on an analysis by the Child Poverty Action Lab, a Dallas-based nonprofit focused on improving the lives of kids. The organization released its findings exclusively to The Dallas Morning News.
The data comes from emerging predictive police technology known as risk terrain modeling. The thesis behind the algorithm, developed during the last decade by two Rutgers University professors, is that environmental factors have an outsize influence on crime and that small informed changes to neighborhoods can prevent it.
Dallas police are set to embrace risk terrain modeling this year as part of their overall effort to reduce crime. The new way of thinking about prevention focuses on places, rather than people. It’s in part an answer to a critique that predictive policing carries too much racial and socioeconomic bias.
Chief U. Reneé Hall is scheduled to brief the Dallas City Council’s public safety committee on her crime prevention plan. Hall’s new-look approach relies heavily on data and intensifying existing programs. Adopting risk terrain modeling, which includes a partnership with the lab, is one of the starkest departures from the Dallas Police Department’s status quo.
Hall’s presentation follows the release of a companion plan by Mayor Eric Johnson, who last week unveiled a list of non-law enforcement solutions developed by a committee of community members to help stem crime.
The formal plans, some of which will need funding from City Hall, were born of one of Dallas’ most violent years in recent memory. The city recorded more than 200 murders in 2019 — a 12-year high.
Alan Cohen, CEO of the lab, believes risk terrain modeling will provide police and the community better data to prevent crime without leading to more arrests, a key goal for his nonprofit.
“It’s a tool of empowerment, not just a wonky data thing that happens,” Cohen said.
And yet, the data alone will not be enough. The model only works if police, other agencies and community leaders work together to understand the information and make structural changes in neighborhoods. As part of the partnership, Cohen said, the risk terrain model and its data will be public.
“Risk terrain modeling doesn’t end crime,” Cohen said. “The actions of police and the community, and the interventions they put in place, end crime.”
Three key components make up risk terrain modeling: a physical boundary, historic crime data and environmental features within the boundary. From there, a series of formulas can identify risk factors and places.
The output is similar to what is known and currently practiced in Dallas as hot-spotting, maps that identify geographic crime clusters on the rise. Police typically respond by increasing patrols in those regions.
But there are key differences, saidJoel Caplan, one of the Rutgers researchers behind risk terrain modeling.
“Hot spots tell you where crime is clustering,” he said. “Risk terrain modeling identifies why it’s happening over and over again.”
The analysis looks for patterns and the environmental features where crimes occur most often. It goes beyond identifying standalone bars or payday lenders as locations where crime may happen by zeroing in on a combination of immediate surroundings, such as liquor stores near bus stops and abandoned buildings.
“It’s never just a car wash,” Cohen said. “It’s never just an alley.”
Given Dallas’ massive geography, the lab divided the city into the Police Department’s seven patrol areas. It then separated those sections into thousands of 250-square-foot cells, each of which was assigned a risk score based on the proximity of different factors.
The lab used 2018 aggravated assaults and spatial features of each city block. It then compared high-risk places to 2019 data. The model accurately predicted between 17.5% and 51.8% where gun violence occurred in 2019.
The range varies based on neighborhoods and types of crime the model aims to predict.
For example, the lab’s model accurately forecast 18% of aggravated assaults with a gun in downtown and 45% of individual robberies in northeast Dallas.
The model’s accuracy increases after adding a 400-foot perimeter, or the length of a typical city block. For instance, 48% of all 2019 aggravated assaults with a gun in downtown occurred within the larger perimeter and 73% of all individual robberies in northeast Dallas happened in the expanded area.
Decades of police and crime research help inform which sort of factors — car washes, bus stops, street lamps — to analyze first. Local patterns and context also help. Police are expected to update and add new data to the model regularly to refine the accuracy of the predictions.
“The model is only as good as the data,” Cohen said.
After computers identify high-risk areas, police and community members work together to create “risk narratives.” These are hypotheses on how environmental aspects work together to create the potential for crime. From there, police, other agencies and community members alter the landscape to disrupt crime.
“The person walking the street is going to know best,” Cohen said, stressing that the data and analysis is only the first step.
More than 30 states and 45 countries on six continents have used risk terrain modeling to some degree. The underlying equations and science have been peer-reviewed and studied in 40 different journals, Caplan said.
Atlantic City, N.J., is perhaps the most celebrated case study for risk terrain modeling. In the first year, violent crime dropped 26% compared to the prior year. Robberies fell by 37%, according to a report by Global Partnership for Sustainable Development Data.
One of the high-risk areas Atlantic City police addressed first were neighborhoods with a combination of convenience stores, laundromats and vacant properties in close proximity. Drug deals were prevalent in those areas.
To curb crime, police officers put sign-in sheets inside convenience stores and laundromats to log their visits. They helped owners add security cameras. And they worked with code enforcement to prioritize cleaning or boarding up vacant buildings.
“We’re getting there before the crime does to make sure that now it is not the next hot spot,” Police Chief Henry White Jr. told The Philadelphia Inquirer in 2017.
The Kansas City Police Department began to use risk terrain modeling last year in an effort to reduce violent crime.
In one example, police identified a pattern of robberies and drug deals in an area on a busy boulevard that included a bus stop, pay-as-you-go cell phone shops and liquor stores. Working with the city’s public transit department, a bus stop was relocated.
Crime fell overnight, Capt. Jonas Baughman told The News. Violent crime since April has dropped across Kansas City between 5% and 59% — depending on the region — compared to a two-year average.
Baughman admitted the department has been slow to gain support. It is making a more concerted effort to involve community members, not just other government agencies, to better understand neighborhood patterns.
“It’s tough to try and get a lot of stakeholders to the table and work from the same data set,” he said. But “there is a myriad of opportunities of who can come to the table and help out.”
There is no comprehensive list of cities, police departments or nonprofits that have used risk terrain modeling because the code is open source — available to anyone. However, Dallas is likely to be one of the largest in population and land size to embrace the technology.
Research, especially out of Atlantic City, also makes clear that the model is just one part of a cultural overhaul within police departments. And to achieve the greatest success, it will take more than just the police department.
That gives Larry James, CEO of CitySquare, a nonprofit that assists the city’s working poor, pause.
“I think the city has the ability,” said James, who has been briefed on the lab’s work. “The question is: ‘Are we willing to collaborate that deeply and pay the price for intense scrutiny?’ This is complex and requires a level of cooperation that we may not have seen in the city.”
There have been similar, albeit smaller, efforts that rely on multiple city agencies to reduce crime. The city attorney’s office regularly coordinates with police, fire and code enforcement to clean up properties. As part of their work, they use research known as crime prevention through environmental design. Small changes such as more and better lighting and keeping bushes tightly trimmed are examples of changes they may require from a property owner.
Jill Haning, who leads the project for the city attorney’s office, said her team could be a model for how the police can work with other city agencies. “We really have a lot of good partners and good people at the table,” Haning said.
The city doesn’t have data on how the cross-coordinated effort is working, she said. However, a study by two professors from the University of Texas at Dallas found enforcing city codes did lead to a reduction in crime.
Supporters of risk terrain modeling acknowledge that change is hard, especially for institutions with long-standing traditions and norms. It likely will be a challenge for Dallas police, which has in the recent past suffered from low morale and waning trust from the public.
“Skepticism can be healthy, but that doesn’t mean the status quo is the best option,” said Caplan, the Rutgers professor. “When these stakeholders are coordinated, pretty amazing things can happen.”