IF DETECTIVES modelled crime patterns in the same way that biologists model the spread of disease, they would be able to work out with far greater accuracy where criminals are likely to strike.
Until now, efforts to predict crimes like burglaries have simply identified 鈥渉otspots鈥� where crimes have occurred some time in the past. The assumption is that if an area has been targeted before, it will be again. But Kate Bowers, a criminologist at University College London, says predictive mapping can be greatly improved if the computer model includes a sense of time, and takes into account how recently crimes occurred as well as where they happened.
鈥淲e have used methods usually used in modelling the spread of a disease to improve crime prediction,鈥� she says. The new crime modelling software her team has developed assumes that, as with an epidemic, houses are more at risk from 鈥渃rime contagion鈥� the closer a previous burglary was both in place and time.
Advertisement
Bowers鈥檚 group tested the software, which runs on a PC, using burglary figures from 1997 for a 22-square-kilometre area of Liverpool in the UK. After analysing both the distances and time lags between crimes committed, they found that an excess of burglaries occurred within two weeks and 400 metres of each other.
One-third of repeat burglaries at the same property happened within four weeks. And houses on the same side of the road as a property where a previous burglary occurred were at higher risk of a repeat crime. 鈥淐ertain criminals are creatures of habit,鈥� says Wilpen Gorr, an expert in predictive crime mapping at Carnegie Mellon University in Pittsburgh, Pennsylvania. 鈥淚f they find a block where they have been successful they tend to go back.鈥� This is because nearby houses tend to have the same interior layout and contain similar consumer goods, he says.
Pitched against the traditional location-based software for modelling crime, Bowers鈥檚 epidemiological model predicted around 20 per cent more burglaries. After training both models using the Liverpool data, the researchers asked them to predict the next two days. The traditional model鈥檚 hotspots included 46 per cent of new burglaries while Bowers鈥檚 pinpointed 62 per cent.
This was not simply because Bowers鈥檚 model produced larger hotspots. When her team calculated the number of crimes predicted per square kilometre of hotspot, its model performed around a third better 鈥� showing that the inclusion of a 鈥渢ime-since-the-crime鈥� element made a real difference to prediction accuracy.
Bowers envisages that police forces will be able to issue officers with new predictive maps each working day. She is now hoping to persuade a British police force to test the system. There鈥檚 certainly interest in doing so. Carl Kruegar, a crime reduction expert with Merseyside Police, welcomes any technology that 鈥渉elps put resources in the right location at the right time鈥�.
