Street crime and fear of street crime have significant adverse impacts on individual lives, the use and regeneration of urban areas, the ability to attract businesses and investment, the price of property, and the ability of citizens to live full and creative lives. Previous studies have examined the relationships between a range of social, economic and situational factors and levels and predictability of crime using a range of techniques. However the impact of altering these factors (where they can be influenced), and how such measures might be combined with other potential crime reduction measures are not fully understood.
This research aims to achieve new insights into the pattern of crime in cities using big data analytics to analyse the relationships between multiple datasets and levels of crime, and use genetic algorithms to derive innovative optimised strategies that result in lower levels of street crime alongside balancing other objectives - such as lower service costs (eg from improved design of street lighting, and policing patterns), lower carbon emissions, and improved public confidence and acceptance. These strategies will be tested through using the city as a living lab, drawing on Glasgow's Technology Strategy Board (TSB) City Demonstrator role.
"The impact of changing street environments on crime: key conclusions from Glasgow
This brief report highlights what seem to us to be the most usable insights that have arisen to date from the ESRC Transformative Research predictive crime project. The main findings are given in headline form first and then explained in more detail. The scope for further work is discussed at the end.
1. Summary of High Level Conclusions
â€¢ The crime rate in Glasgow has fallen significantly, from 33 crimes per hectare in 2004 to 21 crimes per hectare in 2013.
â€¢ Crime rates are elevated around public houses, registered companies and substations; this implies that location has some influence on crime and can assist in crime prediction.
â€¢ There is a wide variation in the amount of crime near to CCTV cameras.
â€¢ 10% of cameras had no street crime within 50m; a review of siting policy may be useful.
â€¢ There is some evidence of crimes which require planning being displaced away from CCTV cameras.
â€¢ There is less displacement at night which may be mitigated by advanced CCTV using laser pointers for example.
â€¢ Good street lighting appears to deter crimes which require planning so improving the lighting may help to reduce crime.
â€¢ Street crime is highly localised and high crime sites are scattered around the city.
â€¢ A significant proportion of street crime arises from a small minority of postcodes so concentrating resources on crime 'hotspots' may assist in reducing crime while also making effective use of resources.
â€¢ Street crime peaks at about 9pm and violent crime peaks around midnight.
â€¢ There is more crime in the summer months and during good weather.
2. Explanation and Clarification of the Conclusions
Prevalence of Crime
During the time period we considered (2004 to 2013 inclusive) the average crime rate for Glasgow was 27 street crimes per hectare. Areas near to public houses experienced higher rates of 106 crimes per hectare while places near to subway stations (55), CCTV cameras (53) and registered businesses (75) also experienced elevated rates.
The crime rate has fallen significantly over this period from 33 crimes per hectare in 2004 to 21 crimes per hectare in 2013.
In general, the crime rate near to CCTV cameras was higher than the Glasgow average. This is presumably because the cameras are sited in predominantly high crime areas.
There was a wide variation in the amount of street crime within 50m of a CCTV camera. There were over 1,000 street crimes within 50m of one of the cameras while there was no recorded street crime near to 40 of the cameras (10%). This raises a question of the siting of cameras for the purpose of detecting crime.
Some indication of crime displacement, from areas of high surveillance, was found for certain types of crime (e.g. assault with intent to rob, attempted murder, attempted theft of a motor vehicle, drug dealing) that are often planned systematically. This insinuates that either the current location of CCTV cameras is inadequate, for these crime types, or that crimes of this type relocate to areas of reduced surveillance. Hence, surveillance could be used as a crime deterrent in certain areas rather than for the primary purpose of observing and catching the offender.
At night, crime displacement was found to reduce for certain crime types (e.g. murder, serious assault, exposure, vandalism), indicating that the use of standard surveillance systems is less likely to deter crime at these locations and the use of advanced CCTV cameras (potentially with laser pointers) might be more beneficial.
From our analysis it can be concluded that there is the potential to improve the location of CCTV cameras so that a greater number of criminal offences occur in the vicinity of a camera diurnally, increasing the likelihood that the accused is detained and thereby reducing the number of future incidents. This may mean a relocation of some CCTV cameras from the city centre to suburban areas, however, due to the high level of light intensity in the area (a deterrent for some types of night crime) and through maintaining the level of spot-checks carried out in the city centre, a change in the level of local criminal activity might not result, and therefore a reduction in wider crime might be possible, at minimal cost (i.e. without the installation of new CCTV cameras).
Overall, nearly a 2% rise in street crime - from the percentage area of Glasgow with a light intensity less than average (8.77 lumens/m2) - was observed in areas of below-average light intensity. Some crime types in particular (related to theft, OLP, murder, indecent assault and public indecency) were more often found to occur in areas with a low light intensity. Hence the potential for reducing the occurrence of these crime types, by improving the light intensity in key areas, therefore exists. Again these crime types, similar to the analysis for CCTV, are often planned systematically and require a heightened sense of the environment.
Looking at the locations of murder, from 2004 until 2013, as an example, the lighting intensity of Glasgow city centre is high (above 54.5 lumens/m2) and consequently there has been no occurrence of this type of criminal incident in the area. This is also likely to be due to the high penetration of CCTV cameras located in the city centre. Murder incidents near to the city centre occur in pockets of darkness where the lighting intensity is between 0 and 8.4 lumens/m2. However this is despite on two occasions in particular, the penetration of CCTV cameras in the local area being high (around 2 or 3 cameras within 100m). Hence, sometimes the level of light intensity counteracts the effect on crime of local surveillance and thus both factors are important in deterring opportunistic outdoor crime.
Patterns of Crime
Street crime displays a highly localised pattern. Out of the 14,733 postcodes in Glasgow, 5% of street crime occurred within the top 22 postcodes; 10% of crimes occurred within the top 68 postcodes. From 2004 until 2013, 1,767 postcodes (12%) had no reported street crimes. This could have implications for policing strategy and suggests targeting resources in particular locations.
There is no clear demarcation between high and low crime areas. Low crime locations occur within generally high crime areas; crime hotspots occur in relatively low crime areas. Nevertheless, the regions are often stable over time and therefore average future crime rates at the postcode level can often be predicted quite accurately.
The prevalence of street crime peaks at about 9pm midweek and at the weekend; violent crime in particular peaks at 1am, with high rates also occurring at 11pm and midnight (midweek and at the weekend). There are very few occurrences of outdoor crime between 5am and 7am. The crime rate is highest during evenings at the weekend.
Weather and Natural Lighting
Crime rates were generally higher in the summer; a 27% reduction was observed in overall outdoor crime, from 2004 until 2013, between summer peak (July) and winter minimum (December) and the bulk increase/decrease in outdoor crime occurs around the daylight saving time dates for the UK (i.e. around the last Sunday in March and October respectively). This increase could mainly be the result of increased footfall in the evenings, brought about by increased daylight and an improvement in weather during the summer season.
From our analysis into the effect of weather on outdoor crime, it can be concluded that overall, crime is generally more likely to occur when there's no rainfall and less likely to occur during periods of light rain, moderate rain and snow. Occurrences of serious assault and drug possession reduce when the level of precipitation increases, however, occurrences of vandalism, theft and robbery increase; where arguably these offences will be carried out in areas, and during times, of low footfall. When analysing air temperature, it was found that overall outdoor crime was less likely to occur when the air temperature was less than 10°C, and more likely to occur above this temperature. Analysing for specific crime types, it was found that criminal offences such as theft of a pedal cycle, the consumption of alcohol (in a public place) and littering were the main positive correlations of crime for increasing air temperatures. Criminal offences, expected to be more associated with a low footfall, such as assault and serious assault were the main observed negative correlations of crime for increasing air temperatures. Assault was equivalent to 18.6%, 11.9% and 10.2% of overall crime during periods where the air temperature was again between -5°C and -10°C, 10°C to 15°C and 25°C to 30°C. Serious Assault amounted to 3.6%, 2.2% and 0.7% of overall crime during the same air temperatures.
3. Addressing further research questions
As part of this project, a range of further questions and research opportunities have been identified, some which may be addressed through further investigation of the datasets brought together as part of the project, or through additional data that, if made accessible, would extend the analysis here. We have divided these into three section outlining what questions could be considered and if appropriate the additional data that we have identified for use to help future analysis.
1) Some issues to be addressed through the current combination of data?
By bringing together the big data in this project - on crime and the street environment - and through the analysis undertaken, several possible areas of further exploration have been identified:
Investigating the characteristics of crime 'hotspots': Is there any difference, in terms of the spatial characteristics e.g. CCTV and lighting between high and low crime locations? What type of lighting is used at the hotspots? Are any of the highest crime areas near to a CCTV camera? How have the hotspots changed over time? Are different crime types associated with different hotspots i.e. is there a robbery hotspot which is separate from a burglary hotspot? As crime has fallen, are the hotspots more or less concentrated i.e. do they account for a larger or smaller proportion of street crime?
Using Spatial Modelling - Geographically weighted regression is similar to linear regression but it allows for spatial variation by estimating different coefficients for different locations. Our analysis suggests that it would be possible to investigate whether geographically weighted regression improves our ability to predict street crime.
2) Adding new data on incidents and victim/perpetrators?
The focus of this analysis, reflecting the ethical clearance provided to the research team from Police Scotland and Community Safety Glasgow, has been on crime events. A range of other data are collected by these bodies and access to them may assist further in the next stage of this research.
Adding crime incidents: for this study only recorded crime (by the police) has been included. Comparable data exists relating to incidents; ie records of calls to the police. This distinction is potentially important in shaping the predictive value of the models. Does an increase in incidents at a particular location precede crime so that it can improve our ability to predict it? How similar is the distribution of incidents to the distribution of recorded crime?
Using anonymised incident reports: data on the home locations of victims and perpetrators would enable the modelling to add some key numeric and geographical aspects or the analysis. It would allow the addition of an analytic determine on how many people are involved in the crime hotspots and how far do they travel? This would have implications for targeting particular individuals and locations and extending the model to separate impacts on victims and perpetrators.
3) Extending the theoretical basis
Several options have been identified. First, applying the theory of complex systems to street crime would bring a new perspective to this problem. Instead of focusing on the characteristics of individuals or the environment, the emphasis is on the interaction between individuals. This is theorised as producing emergent properties such as a feedback loop that generates crime. As a result 'fighting crime' is involved with combating the emergent properties rather than individuals or social conditions.
Second, the analytic and decision making approach here could be extended to investigate particular strategies for reducing crime. What effect does altering the characteristics of the street environment (e.g. controlling access, installing CCTV or other sensors, using intelligent street lighting) have on particular crime at particular locations and on specific groups of (potential) victims? What are the social dynamics of hotspot locations? Why do people go there?"