So.... they're using the information predicted by the model.... to establish the model.

Great thinking, boys.

Odd that it actually seems to be working and has cut crime by 27%... that cannot last.

I guess this just goes to show what happens when you cut the wrong stuff.

If it helps them catch more criminals, I don't see the harm. The police are pretty good at keeping stuff to themselves (anyone remember COFEE?). You don't just need the model to perform an analysis, you also need the stats (which is far from "reading the paper").

Maybe. We will see.

My guess is police use this to identify hotspots that aren't immediately obvious, and emerging trends so they can diverge them. There's no blind adherence, and they're just as capable as the next guy at realising the problems with such a model.

Weren't the police all ready doing this with out the use of a computer?  Maybe all those 'Roll Calls' that we see on those police shows never took place.  'It's a hot summer night out there so be aware of that and check the parks out', 'Oh, and be careful out there.'  I really liked Hill Street Blues. 

No harm, but just pointing up an inherent bias which I hope they're aware of.... hey, me likes less crime.

I like effective police, and any technology that helps the police become more effective. My grandfather was a state trooper, and way too often the police forces get passed over in favor of the big men with big guns and little accountability.

The problem is that, as you pointed out in the original post, their statistical methodology is inherently flawed. They're using predictions to police certain areas more heavily, which will of course result in more crimes caught in those areas, and when that information is entered into the model it will of course assume that criminal activity has gone up there. Since this is apparently part of a system of budget cuts, that means cops are going to be pulled from other areas.

Here's a recent related article on quantum computing which relates to this.  It's about the ability of a computer to determine quantum probabilities and provide accurate answers: http://www.sciencedaily.com/releases/2011/07/110731170028.htm:

And for a fun look...semi-related: http://www.sciencedaily.com/releases/2008/05/080505072755.htm

Simple lesson here with both of these is , "Just because a computer says it's true doesn't make it so."

I read the Times piece behind the b*og blurb you linked.

The data input isn't biased, it's the full database of reported crimes. It will miss unreported crimes, but it doesn't seem like there's potential for a given police force to go 'crime shopping' until they have the day's prediction report, and every force has to make decisions about allocating resources regardless of fancy software. 

Re crooks beating them at their own game, if a criminal group can manage to collect all the relevant data and build their own predictive model, odds are they aren't the kind of group who could get caught with 'predictive policing' in the first place. The crooks cited in the Times article were classically 'petty.'

As far as this tech leading to fewer cops on the beat, pish and tosh. Those creepy camera networks are far more likely to do that sort of harm to community policing. The cops in the Santa Cruz experiment only work those 'hot spots' when they're not on a call.

Computers, even the best, are still constrained by what is fed into them.  But if you feed the correct data in, at least it will crunch it a lot faster than humans can.  How does that help?  Well, if there is a neighborhood that is high crime, you do not need a computer to tell you that.  However, it can help with the infrequent crimes.

The problem with this system is that the cops are both stopping crime and reporting crime. It is thus impossible to measure crime rates without affecting them.

I love seeing anyone point out too-small sample problems, but I just don't get what you are saying about the reporting. Would you have the database exclude arrests made by beat cops who went to an area because of a 'hot spot' report?

That seems counterproductive, unless you're talking about a notoriously corrupt police force such as LAPD and NYPD have been many times in the past. If the cops in Santa Cruz are rotten, those silly Banana Slugs at UC Santa Cruz are not doing their jobs as uppity students.

Exactly. Like rape (under reported), and larceny by telephone/email, etc. (shame preventing reporting).

They don't have to look at 'all crimes'. Just the type they wish to perform and the area they want to perpetrate it in.. The timing will become obvious.

I'm not really sure what the bugaboo is here. This is taking the place of a crime statistician who is using internal crime maps to tell cops where they should be adding more patrols and what crimes to look for. People should be focusing on how this data is interpreted by the cops, which isn't really a technology issue I think. All this can lead to is more responsive (or perhaps spatisically responsive) change of police presence. Real-time beat scheduling, if you will.

Because I honestly don't think you need to tell an experienced beat cop that x crime is likely to happen in y bad neighborhood. Their response would probably be "no shit." This might just add some more finely tuned information, like it's most likely to occur between 2am and 5am. Something a $5,000 contracted crime statistician could have told them after a 2 month survey.

This assumes that policing has no effect on criminal activity, which ain't quite right. Except, again, in cases where cops are corrupt and will somehow keep up arrests even though they've worked neighborhood X or parking garage Y so much that only truly idiotic or drug-addled crooks keep working them.

If that sort of 'eternal high-crime area' happens, it ain't the cops' fault. It's the community's failure to address chronic problems in a specific population--something like a lack of homeless shelters or methadone clinics. Successful professional crooks avoid the cops, or buy them off if they have to.

Nenjin might be overestimating the skill level of the average experienced beat cop, but I agree with the general point about this software basically being yet another form of 'expert in a can.' Now that's a whole different problem for the political economy, especially for those of us who are more or less what Microsoft calls "knowledge workers." 

I've watched The Wire, so I know exactly how those figures are lower....