Skip main navigation

£199.99 £139.99 for one year of Unlimited learning. Offer ends on 28 February 2023 at 23:59 (UTC). T&Cs apply

Find out more

Predicting crime using street structure

Predicting crime using street structure (8:15)
One thing that you’ll notice about these figures is that they look a lot like figures from academic papers, and not a lot like urban areas. And they’re not particularly reflective of what’s really happening. This is just a continuous 2D media with nothing much else going on. So our interest is really in incorporating realistic urban form and what that means for models like this. And in particular, the role of the street network in doing this. So why should we be interested in the street network and the effects coming from it? Well, firstly in the case of something like burglary, for which it’s particularly obvious, targets are arranged on this street network. So everything’s got an address. That’s where the targets are.
Also offenders move around the street network. So that’s how they access the targets, and that’s how they escape from them, for example. But thirdly, we have this idea where, if we just think about risk moving through space, we get counter-intuitive ideas about where risk should go. So just to take a playful example, here’s just a piece of map that I cut out of Newcastle Upon Tyne. And suppose we had a crime happening there. Any crime analyst in the classical British mould would draw a circle around that and say, that’s where crime’s going to spread out. But if you do that, you’ll send some police to the sea. And no burglary’s happen in the sea, right?
But even apart from that flippant example, if we have a crime happening there, say– if we draw a circle around that, then OK, that’s more realistic in terms of what the urban area is like. But we have it defusing over to this street there. So actually that street, it’s close in the Euclidean sense to where the incident happened, but it’s actually miles away if you actually want to travel there. You’ve got to go all the way, literally around the houses to get there. So why should risk communicate over that space? And indeed it almost certainly doesn’t.
So to think about some of the theory of why that doesn’t, here’s a brief introduction to environmental criminology and the theory of why these things matter. So crime pattern theory is kind of the predominant theory of this kind of thing. And what that states is that if we’re thinking about how offenders decide where to commit crimes– they commit crimes in places that they know of, that they have awareness of. And the way that they gain awareness is during their daily activities, which may well be of a non-criminal nature. So they have certain activity nodes. They’re walking around, travelling between these nodes.
And as they do that, they form an awareness space which is the kind of places that they’re familiar with and that they might have some idea about the criminal opportunities there. And of course, all of this is down roads. That’s how people are moving around. And some of that is going to intersect where there are genuine opportunities for crime. So things where some sort of rational actor will think, yes, I will commit a crime here. So by thinking about where these awareness spaces happen, we can think about where people will tend to be aware of crime and where they will therefore offend. This duality between awareness space and criminal intensity.
And aggregating this up across multiple offenders, the kind of obvious corollary of that is that places or streets that are in more people’s awareness spaces, so that more people travel along habitually, will experience more crime. OK. So we can easily start to think about analysing the networks in these terms and see if we can get some results out of them. So it’s trivial to represent a street network as a graph.
There are multiple ways to do it, but here I’ll talk about what is known as the primal representation where you simply place nodes at all of the junctions of the graphs and that’s anywhere where there’s a choice of route, and then connect up any two junctions where there’s a street leading from one to the other. So that’s just exactly the way that you would naively encode a graph. And what can we measure about things like this? Well, we can measure loads of stuff, but the most immediately obvious thing to think about, and particularly given this kind of patent theory idea is Betweenness.
So thanks to Neil for teeing me up nicely by referring to that in his talk a minute ago. I’m going to patronise you slightly more by showing a video of how Betweenness builds up on the street network like that. So the idea of Betweenness– and this may be tedious, but in case anyone doesn’t know, the idea is you take every possible pair of origins and destinations, find the shortest path through the network between those things. And every time a link features– so I’m talking about a link Betweenness here, rather a node Betweenness for those who care.
Every time a link features in a trip, you increment its score by one and the reddest link in the graph is the most between. And it’s kind of picked out the most central there. There were quite a lot people yesterday making a very good point that shortest paths aren’t necessarily the best way to think about travel in an urban area. That’s absolutely true and I make no bones about that. This is just a first order way to kind of approximate these ideas. So if we draw Betweenness for a realistic urban area– this is the city of Birmingham with street segments coloured according to Betweenness.
You can see on the right that it kind of picks out what you’d intuitively think were the busiest streets. And as a point of interest, this doesn’t match up in any way sensibly with the ordinance survey classification, so the official classification of these roads. You get roads which are called major roads, and there’s a huge variation in how you would estimate those to be used across their length, partly because they are very long and a classification applies to an entire road, rather than a street segment. So does this have a relationship with crime?
As much as there’s a perverse appeal in setting up results which turn out not to exist, I wouldn’t have brought you here if there wasn’t something in this. So this is looking at police calls for service in Camden.
That is exactly what it says on the tin– where the police are dispatched, according to 999 calls. And one method which was proposed for kind of looking at urban activity versus street network properties was following this paper by Porter et al, where the idea is simply to compute KDE surfaces of these two things and then look at the correlation between the two. So on the right is the KDE surface of Betweenness. On the left it’s the calls for service. The blotchy artefact is because the nature of the data. And we have kind of done some work on re-scrambling that. Anyway. You can kind of identify some points of interest where it looks like there might be something in this.
And indeed when we do the correlation, these are relatively good, I would say. And they certainly tally up quite nicely with what’s shown by this approach in the other context, which is of economic activity in cities, in the case of Bologna and Barcelona. So these are showing that Betweenness, is in this case, quite a good predictor of these activities. On the other hand, we can look specifically at the street segments. So take street segments as the unit of analysis. And in this case this is taking all street segments of Birmingham. Ordering them in increasing order of Betweenness and taking a moving average of burglary rate as we move through those.
And there’s a reasonably pleasing trend through that where it’s round about 40% more crime is happening at the more between end of the spectrum than at the lower end. And that all stands up with very large significance if you run proper statistical models on that as well.

We’ve talked about how crime can spread through time and space. However, we know that the time it can take to get from one point to another in a city is not just down to the distance between these points, at least not when measured as the crow flies. The street connections between those two points have a substantial influence on this timing.

In this video, Toby Davies explains how taking the street networks of a city into account can greatly improve our understanding of how crime spreads. He also gives an excellent and accessible introduction to a very important concept in understanding all sorts of networks: “betweenness”.

Toby Davies is a research associate in the Department of Civil, Environmental and Geomatic Engineering at University College London. He studies crime using approaches from complexity science, addressing problems such as the influence of street networks on crime, and the evolution of large-scale riots. He has collaborated with a number of police forces throughout the UK, and his work has been featured on the BBC.

You can watch the whole of Toby’s presentation “Incorporating street network effects in models of crime” on YouTube (21:49).

This article is from the free online

Big Data: Measuring And Predicting Human Behaviour

Created by
FutureLearn - Learning For Life

Our purpose is to transform access to education.

We offer a diverse selection of courses from leading universities and cultural institutions from around the world. These are delivered one step at a time, and are accessible on mobile, tablet and desktop, so you can fit learning around your life.

We believe learning should be an enjoyable, social experience, so our courses offer the opportunity to discuss what you’re learning with others as you go, helping you make fresh discoveries and form new ideas.
You can unlock new opportunities with unlimited access to hundreds of online short courses for a year by subscribing to our Unlimited package. Build your knowledge with top universities and organisations.

Learn more about how FutureLearn is transforming access to education