The London riots of 2011


Today saw our new paper on the London Riots published as part of the nature series in Scientific Reports. The work began a few months after the events of summer 2011 – in which London and England experienced the worst period of sustained rioting, violence, arson seen in Great Britain for over 20 years.

Once things had calmed down, two key questions about the events remained: why were some areas of the cities so much more susceptible than others, and second, was there anything the police could have done to bring about a swifter resolution to the unrest? Motivated by these issues, we wanted to create a mathematical replica, or simulation of the events, capable of answering some of these questions.

Our research group at UCL (lead by Sir Alan Wilson in CASA) teamed up with UCL department of Crime Science and the Metropolitan Police, who kindly gave us access to their records of all arrests made in connection with the events. By going through the data, we were able to analyse the patterns seen across London and to create a simulation of rioting across the city by draw analogies with three well known and well studied models from across the human and natural world.

First we simulate the initial decision of a resident to riot. The data shows that events escalated on each day of the riots to a peak in the evening, calming down in the small hours as police began to regain control of the city. Graphs showing this rioter involvement over time are reminiscent of similar plots of seasonal flu outbreaks from year to year, the recycling of various fashions over time, or the spread of a rumour/meme on the internet. By drawing an analogy with ‘the idea to riot’ and these feedback processes (the more people with flu, the more people will be exposed to flu etc) we can simulate the spread of the idea to riot across the city. We took into account that people would be less likely to be involved if there was a heavy police presence (or more specifically, a high chance of being arrested) and, since it was found in the data that rioters tended to come from some of the most deprived areas of the city, assumed that residents in deprived areas were probabilistically more susceptible to rioting

Once a resident in the simulation has decided to riot, it remains for them to choose where to offend. The data showed that rioters tended to choose retail centres relatively close to where they lived, and our model reflects this finding. Perhaps unsurprisingly, this pattern is also found in data on where people choose to shop, and so our simulation takes the form of a traditional retail model – people tend to prefer shopping locally, but are prepared to travel further for a larger retail centre, and the same appears to be true of rioters. This analogy also seems to support the rather flippant headlines seen after the events of ‘shopping with violence’.

Finally, once a simulated rioter is involved in a riot, they begin to interact with the police. We assume that rioters are generally trying to evade arrest, while police are generally trying to reduce rioter numbers, which is reminiscent of the motivations of predators and prey in the wild (officially, Lotka-Volterra models of population dynamics)  –  a very old and well studied problem in mathematics. This forms the final stage of our model, allowing us to test the rioters reaction to (and hence effectiveness of) different policing strategies which theoretically could be used in future to assist the police in bringing about a swifter resolution to any unrest.

The map above is one example of our results (b) compared to the full events themselves (a). Our aim is not to replicate the full riots blow by blow (there is far too much dependence on ‘initial conditions’ for that) but regardless, we get very good qualitative agreement, with 26 of the 33 boroughs showing rioter percentages in the same or adjacent bands as the data.

You can find the paper in all of its delicious glory right here:

While links to deprivation and government cuts to youth services were not the subject or focus of the Scientific Reports paper, this connection did lead me to set up a splinter project, the result of which can be seen below. Happily, it also contains a rather thorough explanation of the model used in the paper which some readers might find useful:

I should point out that the motivations and conclusions of this piece differ from the focus of the paper – some of the work in this video has not been peer reviewed and does not necessarily represent the views of the other authors etc etc etc. blah blah blah. For more info on this project, see my other blog post here.

About Hannah Fry