麻豆传媒

WHY does crime occur where it does? If you thought getting an answer to that question would mean getting inside the mind of the criminal, then you haven鈥檛 met Kate Bowers. She is less interested in the psychology of criminality than in the mathematics of crime itself, and is convinced that this can shed light on patterns of crime. Take burglary at people鈥檚 homes. Although the pattern of burglaries in a neighbourhood may appear random, Bowers and her colleagues at the Jill Dando Institute of Crime Science, University College London, have found through computer simulations that burglaries actually spread in a predictable way, similar to a contagious disease.

It may seem odd to look at the crime rather than the perpetrator, but Bowers believes crime often has less to do with deviant psychology than with utterly mundane human habits and the physical environment in which potential criminals find themselves. In her view, crime is a normal, if undesirable, outcome of ordinary social interactions.

This idea, known as routine activity theory, has been around for almost three decades. 鈥淚t was considered kind of nutty when it was first proposed,鈥� says Marcus Felson of Rutgers University in New Jersey, who dreamed it up. Although still controversial, the theory is becoming increasingly influential as an explanation of how crime works, and in policing communities. Now Bowers and others are taking it to another level by showing how computational analysis can reveal patterns in criminal activity, and even predict where it might happen next.

鈥淐omputational analysis can reveal patterns in criminal activity鈥�

Much as a biochemical reaction can take place when two suitable molecules meet in the presence of a catalyst, the idea goes, so crime results when potential offenders encounter targets in circumstances conducive to crime. In a classic study in 1994, criminologist Daniel Beavon of the Correctional Service of Canada, working with Paul and Patricia Brantingham of Simon Fraser University in Vancouver, Canada, compared rates of crimes against property on various streets in the city with the amount of traffic on them and the ease of access to them. They found compelling evidence for patterns that routine activity theory would predict: for example, of two similar houses, one on a main road and another on a smaller street requiring more turns to be reached, the former was generally more likely to be burgled. This makes sense, the researchers argued, because burglars spend most of their time on routine, non-criminal activities, and when they commit crimes they are likely to do so in places they are most familiar with.

Since then, criminologists have found countless cases where crime patterns can be interpreted in a similar way. For example, a disproportionate share of burglaries and violent assaults in any city take place near just a few tough bars or fast-food restaurants where potential offenders spend time planning criminal activities, which they then carry out close by.

Understanding the ubiquity of these patterns is transforming the way that police try to counter crime. Increasingly, police forces are realising that paying attention to simple factors which increase the likelihood of crime occurring can actually reap greater dividends than dramatic raids or arrests. For instance, following a wave of domestic appliance thefts from homes in Charlotte, North Carolina, studies by criminologists Ronald Clarke of Rutgers University and Herman Goldstein of the University of Wisconsin, Madison, found that most of the stolen appliances had been sitting in newly built premises that were waiting to be sold. The number of thefts dropped markedly once builders delayed installing the appliances until just before buyers moved in.

Finding solutions to crime is much easier once a trend has been identified, and now mathematically minded criminologists say that computer models based on routine activity theory have the potential to make sense of a far more complex mix of social and physical factors that may influence crime.

One of the earliest studies using this approach was led by Michael Batty of University College London. Since its inception in 1964, the Notting Hill carnival in London has grown to attract more than a million visitors each year. With vast crowds jammed into narrow streets, crime is inevitable 鈥� anything from pickpocketing and shoplifting to violent assault and worse. After three people were murdered at the event in 2000, the Greater London Authority commissioned a review of public safety and asked Batty to create a computer model of people鈥檚 movements in an attempt to identify better crowd-management strategies. 鈥淲e simulated the crowd movement around the parade and the exhibits,鈥� says Batty, 鈥渁nd used the model to test 鈥榳hat if鈥� scenarios for changing the parade route, or closing particular streets.鈥�

This led to the discovery that altering the parade route could significantly reduce the density of the crowds. 鈥淭he circular parade route didn鈥檛 let people easily cross it,鈥� says Batty. 鈥淭his was the problem, as all the events were inside the route.鈥� Enlightened carnival organisers adopted the straighter route suggested by the computer analysis, and subsequent carnivals registered a big drop in both maximum crowd densities and the number of crimes committed.

Bowers鈥檚 finding that burglaries spread like communicable diseases is another example of the power of computer modelling. It first emerged from work with her colleague Shane Johnson, completed four years ago. They studied data from the Merseyside region of the UK, containing information on locations and times of residential burglaries committed over 14 months within an area of about 26 square kilometres. This revealed that, following a given burglary, the likelihood of another was increased for the next two weeks for any house within about 200 metres, though the probability tailed off at greater distances and after that time had elapsed. This pattern of communicability of crime strongly mirrors the patterns that epidemiologists find with diseases that spread from one person to another. In the case of crime, Bowers and Johnson suspect, communicability arises simply because burglars have routines, and after one success they often continue in familiar territory nearby ().

To find out whether this trend was unique to the UK, last year Bowers and an international team of criminologists . They found much the same pattern in 10 different cities in Australia, New Zealand, the Netherlands, the UK and the US. Now, a few police forces in the UK have started to exploit the finding by adding patrols near recently burgled houses, or by installing temporary alarms in houses nearby, especially those with similar points of entry.

Though police have been making maps of crime hotspots for years, their predictions about crimes have been based on the assumption that the future will mimic the past. In an effort to improve on this retrospective approach, Bowers and Johnson are developing what they call prospective hotspot mapping (). Their computer model considers each recent burglary as a potential source of 鈥渋nfection鈥� for future crimes. They can then estimate the likelihood of a subsequent crime in any given location by combining the risk associated with each of the previous burglaries 鈥� with those nearest making the strongest contributions.

The approach seems to work. In a recent study, Bowers and Johnson compared the predictive accuracy of their prospective hotspot maps with those produced by more traditional methods. In a typical area of the English Midlands, where the retrospective technique identified the location of 50 per cent of burglaries one week in advance, the prospective technique captured close to 80 per cent. It also located these crimes more accurately, pinpointing hotspot areas down to half the size found by the standard method. 鈥淚鈥檓 a big fan of this work,鈥� says Gareth Hughes, an inspector with the Greater Manchester Police, which has been since 2006.

Understanding space

Taking a different approach to tackling urban crime patterns, Bill Hillier at University College London is mapping the physical environment in which crimes occur, and has developed a way of seeing cities that he calls 鈥渟pace syntax鈥�. The idea is that cities are ultimately built up from elements of space 鈥� roadways, buildings, parks and so on 鈥� and the way these elements fit together exerts a huge influence over where and how people move around, transact business, and generally live their lives. In several studies, Hillier鈥檚 group has shown how a detailed consideration of space and its influence on people can explain a lot of human interactions, including crime.

In the early 1990s, for example, just a few years after the Maiden Lane housing development in Camden, London, had won awards for its innovative architecture, it became a rat鈥檚 nest of social problems, a transformation that baffled researchers. By examining the layout of the development and comparing its character with that of residential areas around it, Hillier and colleagues found that the low-rise, high-density Maiden Lane design (pictured on page 37) had effectively isolated it from surrounding streets. 鈥淎s a result,鈥� says Hillier, 鈥減eople in the centre rarely encountered any other human activity.鈥� This physical isolation matched residents鈥� reports of feeling cut off and vulnerable, especially after dark.

Hillier realised that, with further computer modelling, this same approach could help explain otherwise unpredictable patterns of crime. Over the past three years, he and his colleagues have been exploring how residential burglaries and street robberies in London link to other factors, such as geometrical features of the road network and the architectural properties of buildings. They began by analysing every road segment in a large borough in terms of its attractiveness as a route for getting from A to B. Each stretch of road was given a 鈥渂etweenness鈥� weighting 鈥� a measure of its usefulness for people moving through the city. The result is a map with bright lines showing paths where one should expect large flows of people, either in vehicles or on foot (see Diagram).

While this showed that street crime tends to cluster on these natural paths as criminals seek targets, the pattern of residential burglary was more confusing. For example, Hillier鈥檚 team found that maps of burglary locations often reveal sharply delineated adjacent areas of high and low burglary rates, for which there is no obvious explanation in terms of factors such as income levels. Closer numerical study of the 100,000 dwellings in the area, however, revealed a strong pattern linked to the physical characteristics of the properties: what matters most, it seems, is the type of dwelling.

In particular, detached houses tend to be at a higher risk of burglary than apartments. Indeed, this reflected a more general pattern, in which the risk to any dwelling increased with its 鈥渆xposure鈥�, that is, the number of its sides not directly connected to neighbouring structures. It also emerged that dwellings in areas of lower-density housing were at greater risk. Simple physical and spatial factors are crucial, Hillier says, yet they often interact in complex ways.

鈥淪imple physical and spatial factors interact in complex ways鈥�

Hillier鈥檚 as-yet unpublished findings could add an extra layer of information to the prospective crime hotspot maps that Bowers is working on. Meanwhile, other researchers are using massive data sets of human activities within cities to develop simulations capable of following the movements and habits of millions of people, in the hope of learning more about how crime and other social outcomes emerge out of normal human interactions. Gabriele Istrate and colleagues at the Los Alamos National Laboratory in New Mexico, for example, think this might ultimately be the best way to understand how different human activities feed together in complex ways to produce opportunities for crime, and to learn through virtual experiments how different measures might succeed in reducing them.

Nobody expects any of this work to put an end to crime, but it should help us understand more about where and how it occurs. On a practical level, such objective modelling techniques are likely to provide anti-crime strategies that genuinely work. 鈥淚f we learn to think this way,鈥� says Felson, 鈥渨e can get a lot of mileage without really trying to catch all the criminals, or even to make everyone good.鈥�