Yellen: Have you had an example in the last few months, since the predictive analytics group was put together, where you predicted a crime would occur in a particular neighborhood, and you were able to get there much quick than without the group? For example, where this person said, based on my data, we're going to have something within the next week in this neighborhood because it was busy at this time last year?
Weis: That's hard to measure, because if we get there ahead of time, you are able to prevent it. When you talk to Brett [Goldstein, Director of Predictive Analytics], though, where he can cover you, is where he has anticipated some areas by a block, and like three hours afterward, there's been a shooting two hours away from where he actually thought there would be something.
The fact is you cannot predict random events. Most crime is opportunistic and, yes, random. You might be able to spot trends or similar crimes over a period of time ranging from days to weeks to months, especially in the cases of serial killers, rapists and some robberies, but cops have been doing that without "predictive analytics" for decades now. Even then, it relies mostly on the criminal falling into habitual behavior (comfortable surroundings, repeated successes, available time to commit the crime and available victims to confront). These aren't "GATTACA" or "Minority Report" type scenarios full of genetic markers and "precogs" dealing with preordained outcomes.
And you have to love how Weis comes right out and says, "That's hard to measure, because if we get there ahead of time, you are able to prevent it." So let's get this straight....
- If we make an arrest, we must have been successful in being on scene;
- If we don't make an arrest, we must have been just as successful because we prevented a crime from occurring