Model Citizenry

Developing a mathematical model for predicting murders in Philadelphia.

by Brian Howard

John Toczek loves a puzzle. And he's about to tackle a big one: predicting murders. No, he's not The Mentalist. This is more like Numb3rs.

John — we've been on a first-name basis since Lower Nazareth Elementary — is, in addition to my oldest friend, an operations researcher.

Operations research may not sound sexy; it focuses on analytics and statistics — determining which data in a gigantic data haystack is most relevant — in order to solve big problems. On Jan. 1, John is rolling out a project he's calling the Analytics X Prize (analyticsx.com). It's a contest to develop a mathematical model for predicting murders in Philadelphia, something the Police Department could use to best deploy its resources.

"Operations research and decision sciences started around World War II," says John. "Right now there are decision scientists in Iraq building models trying to determine where IEDs are being placed. It's a seemingly random problem, but if you look at the history of where they've been placed, there are certain traffic patterns, characteristics of roads, ethnic makeups of areas, etc., that begin to emerge. If you replace IEDs in that scenario with homicides, and Iraq with Philadelphia, mathematically, that's the exact same problem."

A few years back, Netflix tried something similar, offering $1 million for the best movie recommendation model. The Analytics X Prize, John admits, is "embarrassingly small" by comparison — $20 monthly and $100 yearly for the best-performing entries (he's open to backers willing to sweeten the pot). The gains could be much greater.

"The models you build can draw on any data sources you think are relevant — changes in housing prices, assaults, presence of graffiti, proximity to a police station," says John.

He got this idea from responses to a column he writes for trade mag OR/MS Today called "The PuzzlOR" wherein he creates math-intensive brain-teasers. "There's a whole community — engineers, medical doctors, students — that's interested in these types of problems. And they willingly do that in their spare time."

John hopes to present the best models to the Philadelphia Police Department's Research and Planning Bureau. "If some of these models turn out to be extremely accurate, you can get into an arena where you can deploy resources before homicides occur."

Given my interest in sports and politics analysis, I asked whether Nate Silver's work qualifes as decision science.

"If you were to replace Andy Reid with one giant mathematical model that did everything from draft picks and picking plays, that would be more in line with what decision science does," says John. "Maybe that's a good idea for the Analytics X prize for 2011."

For more information on Analytics X, visit analyticsx.com or e-mail john@analyticsx.com.