ICIAC11 Posters

The spatio-temporal signature of crime and disorder around football grounds: a tale of Villa and Wolves

(Winner of 'best poster' at the conference)

Justin Kurland, PhD student, Department of Security and Crime Science, University College London

This paper investigates the impact of football matches on crime patterns around two West Midland football grounds (Aston Villa’s Villa Park and the Molineux Stadium in Wolverhampton). The study uses six years of geo-coded arrest and incident data (2005-2010) that were provided by West Midlands Police and the British Transport Police. Crime and incident ratios are calculated by comparing match days and non-match days for non-overlapping 250 metre buffers that extend to three kilometres in all directions from the grounds. The counts of events for both match and non-match days that fell within areas with elevated ratios are then analysed using permutation models to determine if there are significant differences in counts of events both spatially and temporally. The Kolmogorov-Smirnov model is then used to test whether there were differences in the spatial and temporal distributions of events over these same areas. Explanations for identified differences in counts and distributions for match and non-match days as well as the differences in crime patterns across the two stadiums are considered. Finally the implications of these results for theory and practice are discussed. 

I Predict A Riot

Stephen Simpkin, Research and Analysis Officer, Essex County Council

After the recent riots across the UK, attempts have been made to try and explain which key factors led to rioting in some local authority areas. With a better understanding of these factors we can identify vulnerable low level areas and hope to be more prepared should another trigger event spark a similar scenario. National indicator data is available across the country at local authority level. In addition there are other datasets that can add value (eg, deprivation indices). These were pooled together to create a large data file with topline figures for each local authority on a number of themes including: social cohesion; neighbourhood belonging; crime; acquisitive crime; education; employment; anti-social behaviour; poverty and deprivation; and many more.

A dependent variable was created entitled ‘Riot’ and areas which experienced substantial rioting were clearly defined. Key driver analysis was performed on the dataset in an attempt to determine the aforementioned indicators that may have led to rioting occurring in some local authorities. A number of indicators demonstrated a significant relationship with the ‘Riot’ variable. These indicators could be separated into a few common themes: crime and anti-social behaviour (both actual and perceived); attainment and employment; and geographical deprivation.

Once a set of potential drivers had been identified predictive risk modelling could be attempted in order to retrospectively ‘predict’ which local authority areas were most susceptible to rioting. Of course the events had already taken place, so we were not ‘predicting’ rioting, rather creating a set of rules/formulas from the historic data that can be applied to future cohorts of data.

The first approach used was binary logistic regression. This creates a risk score (0-1) of local authorities experiencing riots. The risk score is generated through a calculation of the input variables that have a statistically significant impact on the ‘riot’ variable. (NB – some LA areas had missing data so only 255 LA’s contributed in the model design). The logistic regression model identified 227 areas with a very low risk score, only one of these experienced any problems (Gloucester). The model identified 28 areas with a medium-to-high risk score, 14 of these experienced rioting (false positive rate of 50%, false negative rate of 7%).

This first attempt is very encouraging and, although there are a number of caveats at this stage, the model warrants further development. If data on the predictor variables is available at a lower geographical level, you can substitute in LSOA figures to create a very local risk score. The logistic formula allows for manipulation – so it is possible to see the effect on an LA risk score if there were to be a change in one of the input variables (eg, “a 5% increase in employment leads to a 10% drop in riot risk score”).

Other modelling approaches were investigated on the same dataset (eg, CHAID analysis identified a few key pathways) and as further evidence becomes available on the rioting a robust model is a real possibility. This will allow areas to monitor performance on a number of the key input variables/pathways in order to be continually aware of the risk they could be facing.

An article relating to the poster is published in the Safer Communities Journal 11.2 (2012).  Please contact the presenter for more details: Stephen.Simpkin@essex.gov.uk

Targeting Home Fire Risk Assessments

Julie Luckman, Strategic Analytical Partnership Co-ordinator, Greater Manchester Against Crime

The Greater Manchester Fire and Rescue Service in Oldham approached GMAC to discuss the possibility of designing a model for intelligently targeting Home Fire Risk Assessments (HFRAs) based on the likelihood of where accidental dwelling fires would occur. The overall aim of this was to move away from quantative HFRA targets to something that would ensure the “riskiest” properties in Oldham were being identified. Therefore, the aim was to identify properties where the risk of an accidental fire was higher than normal, and to design a model that would allow these properties to be targeted for HFRA visits.

The work makes use of Experian MOSIAC Household Dataset to enable better targeting of HFRAs. Using this dataset, a profile of the Types in Oldham was constructed. Following this the records of addresses which had experienced a fire were extracted and a second profile of this population was constructed. By comparing the makeup of this smaller population to the general Oldham population it was possible to identify the Types that were over represented. If a Type was over represented in the population of houses which had a fire, then that Type was deemed to demonstrate increased risk. The level to which it was over represented indicated the level of that risk. Once the “risky” Mosaic Types had been identified, different options for translating this into a targeting method were explored.

Mathematical modelling of the efficiency of a variety of targeting methods was undertaken, the following 2 methods were identified as improving efficiency:

  1. Ranking properties in a list according to their Risk value, and then targeting the specific addresses in that order was the most efficient way of addressing risk. It was also the only model that did not require a visit to all properties to eliminate all risk.

  2. A combination of hotspotting methods, which involved a stage approach to targeting geographic areas was the next most efficient method. This involved targeting the following geographic areas in this order:

    1. Hotspots where there had been previous fires;
    2. Hotspots of all properties weighted by their risk score that had not been visited in the previous stage;
    3. Hotspots of all properties (unweighted), that had not been visited in the previous stages
    4. All other properties.

In Oldham method 1 is currently being undertaken and the riskiest Mosaic group is being targeted for HFRA visits.

The future vision for this work is that the analysis and implementation is rolled out across all other GMFRS boundaries by the end of June 2010.

Work is now underway to allow this to happen, and 12 months worth of domestic dwelling fires across Greater Manchester are currently being matched to the Mosaic dataset. Following this the Mosaic Risk Profile will be replicated for all GMFRS boroughs and the riskiest addresses will be shared with the borough commanders. It is important to do this work for each borough individually due to the differences in local population makeup and because of local influences that affect the risk levels of individual Types.

Geographical influences on the night-time economy of urban centres

Victoria Gibson, School of Built and Natural Environment, Northumbria University

This research was carried out with the aim of identifying a definitive ‘tipping point’ at which a town or city’s night time economy (NTE) becomes problematic in terms of violent crime and disorder, and specifically the geographic relationship between NTE centres. The author recognised a paucity of information regarding the wider geographic nature of the night time economy and how the majority of research is based upon temporal patterns and offender behaviour on a localised scale. From analysing the work of others, a relatively crucial research gap was evident - the identification of a ‘tipping point’ which would allow policy makers to proactively manage crime in a town or cities night time economy. The results demonstrated useful information to assist multi-agency activity and policy making in the night time economy.

The author used crime, social and geographic data and systematically selected 28 towns and cities across England to be analysed. A number of statistical tests were carried out, predominantly using linear regression to find the relationships between variables related to the selected urban centres. When patterns started to emerge, the anomalies were extracted and further analysis was carried out to identify external factors which could influence the unusual results.

After considerable analysis and more localised research, it became apparent that the problem of violence in urban centres is significantly affected by surrounding or neighbouring night time economies, in particular the distance between one urban centre’s NTE and its nearest largest neighbour. Analysis was undertaken to clarify and define a ‘tipping point’ distance between such centres at which crime and disorder in the smaller NTE can be expected to increase significantly dependent on action taken within the larger centre. The analysis identifies a combination of influencing factors to explain the nature of the night time economy zones.

The results from this initial study indicate that when a town and its neighbouring larger NTE become more than 18 Kilometres distant, violent crime & disorder in the smaller NTE will increase significantly.

This result gives an indication of the levels of crime per licensed premise as a result of the distance between urban centres and the size of respective NTE’s. Resulting models also indicate that policy & enforcement activity in the larger NTE such as granting of new licences, closing of premises etc will have a significant effect on the smaller NTE. This result suggests a need for Local Authority Licensing Departments and Police Forces to maintain awareness of such activity in their neighbouring NTE centres and pro-actively share information. Current legislation does not allow for activity in neighbouring NTE’s to be taken into account by Policy makers but this research strongly suggests this to be an important factor. This initial research utilised open source data and steps are in hand to extend the analysis by utilising improved data sources and analysis techniques.

Page last modified on 15 jan 12 12:19