Tuesday, August 16, 2011

The LAPD's Department of Pre-Crime

Given the uglier moments in its history (e.g., the Rodney King beating, the Rampart scandal), few police departments operate under more scrutiny than the one in Los Angeles. As such, the LA Weekly asks this interesting question: who polices the police there? The somewhat surprising answer? A number-crunching shop known as TEAMS II.

Would the precogs have foreseen all of Rafael Perez's corrupt activities?
Housed in the IT Division of the downtown Police Administration Building, TEAMS II compiles and crunches data on numbers of stops, pursuits, collisions, claims, lawsuits, complaints, investigations for improper use of force, and instances of discipline associated with each LAPD officer. These data form the basis for a kind of early-warning system, by which the department hopes to identify problem officers before they're embroiled in Rampart-like behavior. According to director Maggie Goodrich, "All that data is fed into the system at the end of the night. Each employee is put into a peer group, so that they're being compared against officers doing similar jobs. There are several thresholds that are calculated by the system." If an officer engages in any of five risk activities, the system determines whether or not he or she falls outside the norm for other officers in the peer group. If so, the officer is flagged and the supervisor notified.

While we're obviously not opposed to greater oversight at the LAPD, the precedent being set by TEAMS II could be a troubling one. According to a VP at the software firm that designed the program, TEAMS II puts the LAPD on the cutting edge of what the article calls "predictive policing". "It uses technology to deploy resources to take corrective actions before something happens ... extrapolating statistics to predict the future." Which is something that any civil libertarian should be deeply skeptical of. For one thing, statistical data can only be used to "predict the future" if the underlying relationships being estimated are stable over time. In the case of human behavior, this is never true: people learn, and their responses to stimuli may change if the (measured or unmeasured) covariates around them change. Second, determining that an individual's average score on some metric differs significantly from a group mean tells you nothing about the practical importance of that difference, and typically it fails to consider the importance of the distribution of such scores. If the LAPD plans to expand this system outside the realm of police oversight, Angelenos should be very concerned about how it might be applied, and how they might be treated if they're "flagged".

4 comments:

  1. And, another important point is that this is grading on the curve. If your peers also suck, individual cops get a lot more latitude.

    ReplyDelete
  2. True. Though they'll have thousands and thousands of data points gathering information this way; even slight deviations from the average will be statistically significant. And it's hard to say where the policy will go, since there's no oversight of TEAMS II; they might ultimately adopt some sort of "good cop" baseline standard.

    ReplyDelete
  3. If an officer engages in any of five risk activities, the system determines whether or not he or she falls outside the norm for other officers in the peer group. If so, the officer is flagged and the supervisor notified

    Am I the only person to think that this encourages bad behavior? If everybody is bad, then egregious behavior is not "outside the norm" and therefore will not be flagged.

    ReplyDelete
  4. @Randian: That's why I raise the point about statistical data predicting the future: people adjust their behavior as they how to advance their own interests.

    ReplyDelete

Note: Only a member of this blog may post a comment.