The LA Times ran a story on Saturday (link ) describing a wondrous tale of the LAPD's foray into predicting when (and where) crimes will occur using entirely statistical methods. (Obligatory Philip K. Dick reference here.) Essentially, they are disregarding things like personal motivation and character and moving forward with the assumption that potential criminals are "rational decision-makers who commit crimes only when they come across opportunities that meet certain criteria." The criminologists that came up with this scheme are George Mohler and Martin Short from UCLA, a mathematician and a physicist, respectively.
Mohler and Short (working separately) want to use techniques from seismology to predict crimes (like predicting an earthquake from seismic activity). They think this will work because of certain "peculiarities" of certain crimes, such as burglary activity in a certain neighborhood increasing immediately after a successful one occurs there. Mohler's model uses statistics like that to cover an area, while Short's model uses things like the presence of barking dogs or missing porch lights to predict individual likelihood of burglary. In both cases, the models work extremely well when applied to real (historical) crime data.
Now, while I agree that more efficient police distribution is an excellent goal (and their results seem to indicate that many burglaries might have been prevented with this model directing resources) I cannot honestly say that these people are on the right track. For one thing, burglary is the only crime they are attempting to model, because by their own admission other crimes such as rape or murder are extremely complex. It may be that crime prediction only works with burglary because people do not behave rationally when committing other crimes. Furthermore, once the PD is using this model, and assuming not all burglars are illiterate, why would a rational burglar burglarize a recently burgled neighborhood? They know that the heat will be on. If they do so anyway, then they are behaving irrationally and the model is operating on faulty assumptions. No wonder this person's paper has "yet to be published."
One additional glaring issue is the "Minority Report"-style vision of the future where police will arrest criminals before they commit a crime. Fortunately, the head of the LAPD's crime analysis unit, Lt. Sean Malinowski, says not to worry: "We still have a Constitution, and we're still going to be arresting people based on probable cause, not on the probability that they'll commit a crime."
That's not very reassuring, is it?