BOSTON, Mass. — The best crime-fighting tools would, in theory, prevent illegal activity from ever happening. Some police departments have adopted tools they hoped would accomplish this. They analyze data on past crimes with an eye toward predicting where more crime might occur. Then they send extra officers to patrol such places. But there’s a risk to that approach: The data used may not reflect the true rates of local crime. And this could threaten to derail efforts at effective crime prevention, experts now contend.
Such faulty data also risk reinforcing racist attitudes and stereotypes, statisticians said at a forum, here, on February 19. The program was held at the annual meeting of the American Association for the Advancement of Science.
Police departments want to do a good job at fighting crime. But the data they work from may be flawed. For instance, those data may not reflect the full range of people committing crimes. Or they may focus on certain neighborhoods while ignoring others where illegal activities also occur.
Such biased data can inappropriately suggest that some areas as especially crime-ridden. Or if certain types of criminals are less likely to be caught — such as wealthy or well-educated people — police might focus their attention on poorer or less-educated individuals.
Kristian Lum is a statistician, someone who collects and analyzes numbers-based data. She works with the Human Rights Data Analysis Group in San Francisco, Calif. In one of her recent projects, she used a computer to analyze where drug crimes had been occurring in the nearby town of Oakland, Calif. The computer then predicted — or modeled — where police should deploy patrols to head off future drug activity.
This computer model concluded that most drug crimes took place in areas with high numbers of low-income people and where many people were not white. So that’s where more police should go in the future, the model recommended.
In fact, Lum argues, it’s not clear how well this model worked at depicting the situation in Oakland. Those data on drug crimes were biased, she now reports. The problem was not deliberate, she says. Rather, data collectors just missed some criminals and crime sites. So data on them never made it into her model.
Story continues below image
One problem: Many crimes don’t get reported
Past research has suggested that drug abuse and crimes take place at roughly the same rates among different ethnic groups and income levels, Lum says. Poor neighborhoods with many ethnic minorities, though, may get more police attention. So officers may see and collect more crime data from such places.
The police and others may be less likely to notice and report drug crimes in well-to-do areas. Officers also may be less likely to stop someone for suspicious behavior in certain neighborhoods. They might instead focus on where crowds are likely and where people on the streets often interact. If crimes also happen elsewhere, the police may not know about them.
Therefore, if a model is based only on police data, it won’t include all crimes, Lum argues. “To know all crimes, you’d have to live in some sort of surveillance state,” she says. By that, she means the government would watch your every move. And no one wants that.
“You’re going to find things where you’re looking,” Lum says. If you mostly look in inner-city black or Latino neighborhoods, most of the crimes you find will be there. And any computer model working with such skewed data would mistakenly predict that these areas are hot spots for future crime. Such a model would reinforce the idea that certain groups of people are more likely to be criminals — even if they aren’t.
That idea also runs afoul of basic U.S. constitutional law. (A new, February…