Monday, April 16, 2012

Oliver’s Twist: Has the SPD pawn shop sting reduced property crime?

Note: I wrote this a year ago, when I decided to commit to a career in data science. I did it to refresh some skills, and to give potential employers a sense of how I approach problems. It worked really well. If you are a data science newbie, I recommend doing stuff like this. -- K2 -- 2013-04-01

Here’s an example of how easy it can be to explore a data set when you combine powerful, user-friendly tools like Socrata (which publishes public sector data), and the free version of Tableau (a great visualization tool). To try this analysis out yourself, check out the Seattle Socrata data sets here, and the free public version of Tableau, here.

Property crime is Seattle’s biggest crime sector. When you combine vehicle theft, burglary, and larceny, you get 92% of Seattle’s criminal incidents.
“In 2010, Seattle bucked a national trend of declining property crime rates, with burglary and theft rates here increasing 3.2% in contrast to a 1.3% decrease across the country, according to data from the U.S. Department of Justice’s “Crime in the United States” report.” (article here)
image
(Data set: Crime stats by precinct, 2008-2011)

This struck me as rather a lot of property crime for an area with such low levels of violent crime, so I tracked down a local Seattle beat cop to get some perspective on it. He told me about an interesting pawn shop sting the Seattle PD ran over the last year to attempt to address the rising property crime levels, named Operation Oliver’s Twist.

Mayor McGinn described Oliver’s Twist in a press conference on 3/6/2012:
[D]etectives from the Seattle Police Department’s Major Crimes Task Force and the Pawn Shop/Property Recovery Unit, working with the King County Prosecuting Attorney’s Office (KCPAO) and the FBI, set up a storefront fencing operation – a tactic not used by SPD since 1979 – where undercover officers spent 11 months buying stolen goods from suspects for pennies on the dollar, with no questions asked.
With initial results:
As a result of the operation, detectives identified 102 suspects involved in 314 separate criminal cases. Dozens of suspects were arrested and booked into jail over the last 24 hours on their outstanding cases.
The question is: has this operation shown immediate results in reducing Seattle’s property crime?
image
(Data set: Police report incident, used for the rest of this blog entry)
(Data transformations: here, you can see how I’ve defined Seattle property crime.)

Overall, property crime in Seattle hasn’t decreased since the sting arrests began in (assumed) Feb-March of 2012. In fact, from the regression line, you can see the total number of Seattle property crime incidents has actually increased a little.

Let’s break it down by district to see if we can see decreases in crime closest to the sting pawn shop. I used my data set to cobble together a basic Police District map of Seattle. The SPD pawn shop was located in Georgetown, which is in the light pink “O” district below.

image

Now we know that districts F, K, M, O, R and W are closest to the SPD pawn shop sting. We might (theory) expect to see property crime decrease most dramatically in those districts. To see the difference, I compared the number of property crime instances in the months of February and March in 2011, to the same period in 2012 (the estimated time of the arrests), in each district.

image

To make these stats a little clearer, let’s graph the change by district on our original district map. The blue circle describes the neighborhood of Georgetown, the location of the sting. Red indicates an increase in property crime incidents, green a decrease. From this map, we can see that the pawn shop location is surrounded by a lot of red – a lot of districts for whom property crime instances actually increased between 2011 and 2012.

image

So, working from this data alone, it looks like the “Oliver’s Twist” sting in the O police district has not yet caused a decrease in property crime in neighboring districts.

It’s still possible that arrests are still ongoing, and the positive impact of the sting has yet to fully manifest itself. Looking at a rolling average since the beginning of February, we see that there has been a steady decrease in property crime in Seattle over the last few weeks (though levels still remain higher than this time last year). It’s interesting to note that the drop is substantially more pronounced in districts farther away from our Georgetown pawn shop.

image

There are a lot of factors at play when it comes to measuring crime rates. How much might property crime rates have grown without this intervention? How have previous large-scale stings in Seattle and other areas impacted overall crime rates? Is it possible that there is a negative relationship between location and property predation? Perhaps criminals fence stolen goods intentionally far from where they grabbed them, and the “Other Districts” category is really the one that contains our signal. All interesting questions for the next intrepid analyst who wants to play in this space.

Remember: all data sets are quirky, and no analyst infallible. Please have your own team confirm these results before basing any big changes on it.

Studying data science is a lot like being a data scientist

Note: Last year I decided to commit to data science as a career. I did this analysis to brush up on some skills, and to show potential employers how I solve problems. It worked like a charm, and I recommend it to new data science people. - K2 - 2013-04-01

Data science is like anything else: the best way to learn to do it is to do it. This is challenging if you’re winging it, because there isn’t a clear path laid out for newbies. There are lots of free / low cost resources out there, but most of them assume some previous knowledge from the other resources. It’s unclear what comes first, which data philosophy an author / instructor is operating on (there are several), or which techniques are most practical in the real world. Thus, learning data science is a lot like doing data science: you start with some half-formed questions, search and slice until you have some half-formed answers, organize them somehow, refine your questions and start again. Making it work curiosity, and a knack for sorting through giant piles of unsorted information and turning it into categories. The good news is: you’re probably already good at that, which is why you’re interested in data in the first place.

The other good news is that I’m going to lay out some of those steps & terms for you here. Personally, it drives me nuts when things are made to seem harder or more forbidding than they have to be. While data science isn’t for everybody, there are way more people out there who would be great at it, than there are people who know they would. The industry is going to need all of us: the ones who know they can do it and the ones who don’t. So, I figure, let’s lower the bar of entry. If each newbie works to make it easier on the next newbie, before we know it there’s an army of us well-poised to ask and answer fascinating new questions about human behavior.

The flip side is that since I’m just getting oriented myself. Collaboration is the steam that makes data science go: if you want to add a resource, step in the process, or advice to this ground-up tutorial series, let me know.

Next: we start by doing. I’ll set you up with a couple of user friendly tools that let you circumvent some (though not all) of the initial technical hurdles, so you can get directly into the fun part: data analysis.