Friday, August 3, 2012

Sci-fi policing and the Holy Grail

"When predictive experts fail they are just replaced by a new group who say they can do better." John Ralston Saul

The prediction game is making the rounds again. If we are to believe the New York Times and the Huffington Post, there is this tiny corner of criminology where computer scientists and math types squirrel away like mad scientists decrypting secret code.

The predictive experts are the latest media darlings. No facetiousness intended. If prediction delivers on what it promises and becomes an early warning sign to improve community-based, proactive problem solving you can count me in.

IBM and Memphis Police

The latest iteration, this one from IBM, is on a crusade and they have some new friends in police departments like Memphis.  Memphis points to their new predictive policing program called "Blue CRUSH" to account for a 26% crime drop in the past five years.

Sci-fi policing in real-time! Kewl. The mad scientists are positively tingling. IBM's background report explains it all.

Wait! Not so fast.

From 2002 to 2011 Tampa had a 72% felony crime plunge. Police say Tampa did it with proactive problem solving, analysis from their crime mapping unit, and a compstat performance review to hold area managers accountable. Others point to demographics. Tampa had a 40% increase in 50-64 yr olds with the financial resources to trickle some positive mojo into the local economy. That in turn mitigated, or displaced, younger crime-prone age groups.

Either way, predictive policing had nothing to do with it.

Predictive policing and the promise of sci-fi
Still, I have a soft spot for the sci-fi promise. My blogs show it; Solving the city with math and Predictive policing and the PreCog paradox are the latest examples.

Disclaimer: In 1988 I co-published a predictive spatial analysis on probable locations of professional auto thieves. In the 1990s we expanded that into a tipping point theory predicting how neighborhoods tip into crime. Sci-fi policing groupie? Guilty as charged.

All great fun. All beside the point.

Police resources nowadays, razor thin and bloated on salaries, can scarcely afford expensive math experiments. Tampa did fine without it. Would they do better with it? Maybe. But if demographics are the primary cause of crime declines then we're fooling ourselves with fancy math and ignoring the root social causes that trigger it in the first place.

Anyway I'll continue reading about it. After all, who can resist such appealing titles like Poisson-based regression analysis of aggregate crime?

Not me.

2 Replies so far - Add your comment

Tim Hegarty said...

As I said previously, we are looking at one of these probability policing (the more accurate term) software solutions, and we are seriously considering it. The company in question is clearly in the business of making money, but the only one promoting it as the Holy Grail is the media (not unusual). The company itself is promoting it as A tool in the crime reduction tool box, not THE tool.

We have managed to reduce crime (and the crime rate)three straight years without probability policing software, so why would we consider spending a significant amount of money on something that we can already do ourselves? About a decade ago, I learned to forecast trends with a series of mathematical formulas and a TI calculator. Each individual forecast took about 10 steps, a page of legal pad paper, and a lot of double-checking. It was very time-consuming. Buying Microsoft Office now allows me to do the same thing in a fraction of the time with an Excel spreadsheet. What do I gain by spending money that I don't really need to spend? Time. Time to do other things, making me more productive and cost-effective. And frankly, more confidence in the forecast numbers because their not subject to my human error.

Our crime analyst suggests that having probability policing software during the first half of this year would have saved him about six weeks of time. That's time that he could have used assisting with major investigations, or supporting our repeat offender unit, or providing data and analysis to a problem-solving team. Is that worth the money necessary to purchase the probability policing software? Maybe. We are developing a crime intelligence unit, and the initial thought is that it will need three people. With this software, which costs far less on an annual basis than the salary and benefits of a real person, perhaps we only need two. I think that's worthy of consideration.

Would it make us better at accomplishing our mission? I don't know. It's sexy and in the news, and unfortunately, some in our field that are using it are enjoying their 15 minutes of fame a bit too much. I wish PTO was getting the same coverage, or SafeGrowth. Thankfully, neither of these has ever been promoted as the Holy Grail, but we'd sure be missing out on something good if we rejected them just because someone had.

In keeping with the title of your post, I'm sure even Blue Crush doesn't know the air-speed velocity of an unladen swallow.

GSaville said...

Thanks Tim for your excellent points and real-time update.

I agree with your comments. Predictive strategies are fascinating stuff that might smooth out crime responses.

Whenever I see a scrumptious buffet I always have the same problem: Do I chose what's healthy? Do I dive into the chocolate? Do I nibble at everything?

With limited appetite I must choose carefully. I know I need wholesome things that will keep me healthy. But life is also about those cool things that spice it up. And I love chocolate.

There are proven strategies that deep dive into crime: the PTO Police Training Officer program, problem-oriented policing, and SafeGrowth. They have proven track records for cutting crime and improving police quality. They may not be chocolate-sexy, but they work. They keep us healthy.

I hope you are right that "probability" policing will free up time for the deep dives.

In the police organizations I've worked I discover when time is freed up it is rarely the deep dive strategies that benefit. Instead response smoothers get more PR attention and the deep dive stuff gets ignored. I hope I'm wrong. Maybe your organization is different.

Only vigilance by enlightened leaders, people like you, will ensure the money spent on predictive models rolls over into deep dive strategies like PTO, SafeGrowth and POP. Funding one should therefore lead to funding the others.

Ah, the buffet! I just love the chocolate :-)

Incidentally, I've heard it before but I don't buy the "probability" vs prediction semantics. In advanced statistics there is an undecipherable distinction between probability and prediction. True, they diverge, but in many cases they are used interchangeably. I'm unconvinced that simply using the word "probability" convinces any policy-maker that prediction is not the game at hand.

See this example: