Back in December, our flat was burgled. Our area of London, faces a higher burglary rate. Just before our home was hit, several others in the area were attacked. Shortly afterwards, some others down the road got it. Then things in our block went quiet, but there was an uptick not far away. Last week, a few blocks from us our friend’s house was sacked, and another house between there’s and ours had its back-garden shed hit. Then the other day, the house two doors from ours was hit again. The migratory pattern here seems remarkably un-amazing. The fact that there are so few (if any) arrests or convictions is more interesting.
By in large, humans are predictable creatures, we just don’t like to think of ourselves as such. A couple of years back, a study showed how analysis of mobile phone data showed the predictability of how people move. The visualisation showed a sort of ant trail, or bird migration kind of repetitive cycle. The Center for Complex Network Research analysis of the data found that human behavior is 93 percent predictable. What’s invisible to the ground-eye view is much more apparent when sitting from a high vantage point with the correct data displayed in the right way. This works for a lot of things. It evidently works for some types of crime.
People as a species seem overly proud of our “big brain” advantage over the rest of the animal kingdom, but that may be an illusion. Our huge brain spends a lot of its processes trying to find short cuts to doing real work, employing what Sorin Pintilie refers to as psychological tripwires to take on the heavy lifting: “When we’re thinking, what we’re trying to do is to look for the nearest, available pattern. Once we can find it, we can stop thinking and just follow along the pattern. This makes for a very efficient computing machine we call our brain.”
Human predictability upscales very nicely. Nicholas Christakis and James Fowler studied social networks (real world ones, not websites) to create a predictive model about epidemics. This seems to have morphed into research using Twitter data to predict the onset of a flu epidemic. Using what Science Daily describes as Syndromic surveillance of online microblogging data, researchers predicting flu outbreaks on a wider scale “obtained a 95 percent correlation with the national health statistics collected by the CDC. In addition, the results were comparable to figures collected by Google with its Flu Trends service, which tracks influenza rates by analyzing trends in query terms.” Speaking of Google’s Flu Trends: “We’ve found that certain search terms are good indicators of flu activity. Google Flu Trends uses aggregated Google search data to estimate flu activity.” Being able to predict the liklihood of certain types of crimes in time and space works similarly.
In the video above, Emory University researcher Donal Bisanzio created a data visualization of the movement routines of people in Iquitos, Peru.Good data visualisation comes in by not only mapping location, but time. “Spatiotemporal” data processing leads to better predictions when it’s culled and presented in a meaningful way. It works in trending lots of things, from voting trends to advertising.
Crime records are often very comprehensive, so they certainly qualify. I can’t say what police may be using behind the scenes, but it seems demonstratively as reactive as the public view of UK and London crime stats that are published on various websites. Let’s take a look at them:
GLA strategic crime analysis: Numbers are up or down, but not much on where, when or why. It doesn’t get to a borough level, and has no predictive capacity. This tells you very little, really.
London Met Crime Map: The image at left is about as far as you can drill into a trend and region and given period and compare. It’s more detailed, but it would be difficult to see what’s coming or to spot a real trend. You can see some area-by-area information, and the raw data has more of a chance to be used in a better way, but it’s not quite in focus. You wouldn’t divine from this where the robbers are going to hit next, really.
Police.uk maps: Now we’re getting somewhere with data visualisation. Good data visualisation tells a story.It can be artistic, it can also narrate some kinds of events in the future tense with an increasing degree of reliability.
This is where we get into choppy waters. “We should always keep in mind that any new technology that helps the police to better protect citizens can also be used to better oppress them” writes Reason’s Science Correspondent Ronald Bailey. This is demonstratively true, as history repeatedly shows. But it’s too late to pretend this one doesn’t exist. Police in our area, and throughout the city use electronic mapping, but seemingly little more than a Web 2.0 version of putting tacks in a wall map.
The data is now being collected, it’s just put to poor use. This model does away with socially corrosive and grossly inaccurate systems such as racial profiling, which London police have been accused of on more than one occasion. UCLA’s model is based on data interpretation of already established events, devoid of racial overtones: “Predictive policing is based on the idea that some crime is random—but a lot isn’t. For example, home burglaries are relatively predictable. When a house gets robbed, the likelihood of that house or houses near it getting robbed again spikes in the following days.” And unlike racial profiling, or so-called “random” stop-and-search, there’s evidence of effectiveness. I’ll take the data, thanks.
Unfortunately, this isn’t the method that London Police are taking. To be clear: They’re doing it exactly wrong. Instead of using evidence-based predictive software, they’ve brought the simplistic profiling to computers. They use Geotime to track individual suspects movements. This doesn’t really tackle trends or reduce the types of crime that impact a sizable chunk of society, but is applying error-fraught human element of choosing who to track in a more invasive way that’s also subject to being abused by authorities. This leads to people involved in political dissent (not a crime) being monitored, and money being wasted that could be going to make communities actually more safe.
Using something like Geotime may seem logical to people, but it’s backwards: Resources are going into trying to predict the movement of individuals police know the identity of, but are not sure if they’ve done anything. This is instead of tracking crime trends that have been verified with the goal of predicting where they’ll happen next and then seeing who’s doing it. The first one may lead to random one-off successes and making a number of mistakes along the way. The second has evidence in reversing crime trends.
Returning to the premise that humans are essentially predicable creatures, there are elements missing from UK published crime maps: What were the locations of police, street by street, during these time periods? It would be interesting to show a visual movement of police over a given time period in the same map view showing crime movement. My guess is that in both cases a pattern of predictability would be apparent and overlap very little.