Embedded Files
Data and map files used in this guide available to download as GeoPackage from following link: https://www.dropbox.com/s/h0550hivi7dl89l/belfast_hotspots_guide.gpkg?dl=0
Map of City of Belfast admin districts
1. Identify parameters and data
1. Identify parameters and data
The first step in designing a proactive visible policing hotspot initiative is to choose a problem and think about geographic data requirements. I've prepared a data set from police.uk open data covering the city council area of Belfast, Northern Ireland, and excluding the wider urban area that falls outside the city council limits (see map left). I am using four of the broad aggregated crime types available from police.uk - drugs offences, robbery, violent crime and weapons offences. Ideally, you will be working with more granular data that is problem-specific and focuses on a subset of offences that are street-visible, not occurring indoors, for instance. You will also be working with data that has not been subject to geomasking that compromises spatial accuracy (see Tompson et al 2013).
2. Data preparation
2. Data preparation
Variables. The minimum dataset required would include date/time of incident, precise location (coordinates), address/venue information and the crime type classification. Additional fields may be required for assessing content and quality of data, such as summary free text which is useful when checking potential outliers or anomalies that could distort spatial patterns.
Weighted values. If you are planning on creating harm spots using a weighting system, the correct crime type classification is required to use as a lookup - in England and Wales, this is the Home Office Code (Crime Severity Score) and Home Office Classification Code (Cambridge Crime Harm Index). There is a growing body of literature on the use of harm weighted hotspots available in the Cambridge Journal of Evidence-Based Policing (see Weinborn et al 2017), and a useful assessment on the differences between the two methods (see Ashby 2017).
Timeframe. How far back should you look? This varies from study to study with the minimum typically being 12-months of data. The Belfast data I've prepared contains three years worth of geocoded data. The Philadelphia Foot Patrol Experiment identified locations using three years of data, applying a recency weighting (see Ratcliffe et al 2011). It is helpful to develop a good understanding of crime concentration and stability in your area as this may help inform a decision (see Telep & Hibdon 2019). A challenge that often can present is the time-lapse between the identification of study locations and the deployment of operational resources. Ideally, the start of a project would coincide closely with the end of the analysis to take advantage of having the most recent (12-months) data.
In the open data example below, I have added a pseudo harm score for the broad offence classifications. I have also created a score for recency (75 for 2019, 150 for 2020, 225 for 2021), this was in conjunction with harm values. You may use 1, 0.5 and 0.25 if you were using counts, making sure to weight points closer to the present higher than those further away in time. The combined 'HarmRecency' variable can be used to weight hotspots when you begin mapping, if that is your chosen approach.
Data quality. Once you have prepared the data, it is worth exploring to identify any outliers that could distort spatial patterns or produce 'manufactured' hotspots. By manufactured, I mean crimes where the precise location the offence took place may be unknown, and for administrative purposes, another location has been recorded. Common examples of this include police stations, hospitals or health services, courts and educational establishments - one example is where crimes have been disclosed in such settings and then reported by a third party to police. Two possible outliers in the Belfast open data are the Royal Victoria Hospital - there are 422 offences at this location. There are also over 400 offences at the point of Musgrave Police Station and adjacent Court House. More granular or qualitative data would allow us to assess the context and quality of this information. In the absence of more detailed data, I'm making a decision to remove these two points from the analysis.
3. Decide unit of geography
3. Decide unit of geography
Visible hotspot patrols are best utilised in areas small enough (and hot or harmful enough) for a police presence to act as general deterrence. Areas that are too large (such as administrative boundaries or patrol beats) can minimise the effect by diffusing visibility. Sherman and Weisburd (1995) groundbreaking trial found that more geographically concentrated patrol areas could reduce crime and disorder. The most commonly used units of geography applied in hotspots studies are equally sized micro-grids or similarly sized street segments, the latter may require some adjustment when working with street networks that do not follow uniform grid layouts, as is the case in most UK urban areas.
Street segments. Polylines using road and street networks, often divided or intersected at equally sized intervals. This is the preferred geographic unit of choice (see Chainey, 2020, p.42), and requires that you have access to point record level incident data and street polyline data. Street network data is readily accessible for many areas via Open Street Map (OSM). There are a number of ways to access this data. It is possible to download files containing shapefiles with street network polylines for entire regions (http://download.geofabrik.de/), and working within QGIS there is a dedicated easy to use plugin to retrieve OSM data via API (https://plugins.qgis.org/plugins/QuickOSM/).
QuickOSM Plugin outputs shown below to retrieve Belfast street network
QuickOSM in QGIS, select Highway and set for canvas extent to retrieve street network in your study area
Output created as polyline object
Attributes table includes numerous variables, most useful to keep are id numbers, highway classification and street name.
Micro-grids. Polygon shapes of a set-size, often selected are squares of between 125-250m. The simplest method is to create a 'fishnet' of equally sized grids across a city jurisdiction and count the number of crimes (or weighted score) per grid cell. Be mindful that this method can dissect or mask hotspots depending on where lines are drawn, or cause visibility to be applied across an entire grid when it may just be one venue or address that has contributed to the count/weight. An alternative method is to combine the use of micro-grids with another method, such as KDE hotspots (Kernel Density Estimation), whereby centroids (a point at the centre of a graduated hotspot) are used to create grids. The output will look more like a spread of measles.
Other methods. Cluster analysis (i.e., k-Means, Ned Levine algorithm for nearest neighbour clusters, DBSCAN - Density-Based Spatial-Clustering of Applications with Noise) can locate precise geographic boundaries for hot or harm spot clusters. Less preferable options would be to use census or administrative geographical boundaries. In a recent study testing larger areas, UK Census Lower Super Output Areas (LSOA), significant reductions were reported, see Bland et al, 2021.
A consideration when choosing the size of micro-spaces is that problems could emerge when attempting to track officer movements accurately within them depending on the accuracy and effects of GPS refresh rates of police radios (see Hutt et al, 2021). An example might be that, if you choose a micro-space size of 100m and the tracking or GPS being used has a geo-positional error or refresh rate greater than 100m, it could make tracking visits and compliance unreliable. Some forces have purchased trackers specifically to increase reliability tracking police patrol dosage and compliance (see Basford et al 2021, Bland et al 2021).
Examples of mapping methods below. All maps were created using the free and open source QGIS. Base maps source ESRI, DeLorme, HERE, MapMyIndia. Graduated street segments, grid map and nearest neighbour cluster algorithms are from the Visualist Plugin for QGIS (Rossy, Q. 2019).
Graduated street segments, mean segment length 140m, max/mode 200m.
Grid map, 250m x 250m squares
KDE Hotspots 150 bandwidth, grids applied 250m x 250m squares
Nearest Neighbour clusters, variable size, minimum cluster size is 36 offences.
4. Calculate crime concentration
4. Calculate crime concentration
The next step is to identify the hottest or most harmful concentrations of crime. Numerous studies across the world have found that crime is concentrated at a relatively small number of micro-places. Furthermore, long-term time series of crime at micro-spaces has found strong stability in these patterns, known as the 'Law of Crime Concentration' (Weisburd, 2015). Hotspot studies require us to identify these places to implement our intervention and have the greatest potential impact for crime reduction. Calculating crime concentration requires rank-ordering your units of geography in descending order from the highest volume or harm, and calculating the cumulative frequency (running total). This is something you can do within GIS attribute tables, or export and calculate in Microsoft Excel or Google Sheets. We can use the table to see clearly what % of crime is contained within what % of units.
Using the Belfast street segments data, we can see that just 1% of those units (79 street segments) contained 21% of all crimes under study. The number of annual offences within these segments ranged from 19 to 112, with a mean of 36. The top 5% (397 segments) contained 46% of offences ranging from 7 to 112, with a mean of 16. Almost 50% of street segments contained no recorded crimes for the categories under study.
How many areas to be selected for intervention is likely to be dependent on resource capacity. The more areas selected for a study, the more likely it is that there will be a need to include areas with much less crime than others - for example, in the 5% highest in Belfast the range in offences is from 7 to 112 per street segment. As more areas are selected, the greater the variance, which could make it more difficult to detect significant findings (see Weisburd et al 1993). This is considered when we determine how areas will be selected and assigned.
5. Choose areas for intervention, and control groups
5. Choose areas for intervention, and control groups
Control and treatment. It is beneficial to identify areas for treatment (intervention, where visible hotspot policing will occur) and control (comparison areas that will not receive the intervention) so that we have a basis for comparison when it comes to assessing effectiveness. If we observe a significant change in our treatment area comparative to our control area then we can have greater confidence that our intervention has resulted in the effect. There are many different study or research design methods that can be applied to hotspot policing, and they also exist within a hierarchy of evidence (see article 'Not all evidence is created equally'). Two possible methods are described below using the Belfast KDE informed micro-grids data.
Method example 1: Block randomisation
In this example, we selected the top 1% of KDE informed micro-grids in Belfast (66 250m * 250m grids). The average annual volume of crime across these grids ranges from 12-260, with a mean of 57 per year. They contain 27% of offences. We rank the grids in pairs by their level of crime, or harm. As an example, the images show how each grid is paired alongside another grid with a similar volume of crime, or sum of crime harm. For each pair one grid is randomly selected to be assigned to treatment and by default the other is control. The map shows how this might look geographically.
Above, arrows show matched pairs, colour fill denotes treatment (red) and control (purple)
Method example 2: Randomised cross-over design
In this example, we have selected just the highest 30 KDE informed micro-grids in Belfast, based on their volume, or sum of harm. The average annual volume of crime across these grids ranges from 49 - 260, with a mean of 89 per year. They contain 19% of offences. How this method works is that each day 1 in 3 grids, for example, would be randomly assigned to be visited for the visible policing intervention. This is illustrated in the map visual, where on each day of the project a subset of the grids (highlighted in yellow) would be assigned for treatment and the remainder would be control grids for that day.
Randomising. Randomising daily visits is a task that can be completed using a variety of software programmes. A simple solution is provided below for the cross-over method, which can be replicated in Excel. All that is required is the unique names for each micro-space (i.e., grid cell, street segment or another unit of geography), and the number of units that you require to be chosen at random for each day. Another option is to refer to 'The Randomiser' resource for this process. Make sure that the ID/identifier you use for your treatment areas remains consistent across your GIS files, randomiser, any briefing materials provided for officers and within datasets that track treatment dosage. All this information will need to be brought together during the assessment and evaluation.
Considerations. Some other threats to the reliability or validity of a project to be considered are below:
Control groups and treatment areas should be as similar to one another as possible
Control and treatment areas should have a sufficient distance or buffer between them. This is to ensure that control areas are not contaminated by interventions in treatment areas. This can sometimes be challenging, such as when many suitable micro-places identified are adjoining or clustered together in close proximity.
We may wish to include buffer zones to assess local displacement.
Some extremely useful resources to check out which can assist in research design are included in the reference section of this page (see Ratcliffe & Sorg, 2017; Braga et al, 2019; Mitchell, 2019; Bachman & Schutt, 2020; Chainey, 2020; Linton & Ariel, 2020; Chainey et al, 2021).
6. Tracking implementation
6. Tracking implementation
Once the treatment areas are identified and the assignment schedule has been determined, and before the treatments have been administered, a method for tracking dosage and compliance needs to be set up.
Dosage. A study by Koper (1995) suggests that there is an optimum level of visible patrol dosage in suppressing crime in hotspots. It found that intermittent patrols of 10-16 minutes can extend deterrence. Replication studies in the UK (including Belfast, Birmingham, Liverpool, Peterborough, Southend) have aimed to deliver random and intermittent 15-minute patrols in micro-hotspots on assigned treatment days, during peak crime hours. A range of methods exists for collecting data on dosage. These vary in sophistication and are often chosen on the basis of technological and software capabilities available. CAD and Automated Vehicle Locators, officer Radio GPS, purchased hand-held GPS trackers or reliance on officers completing a return form (i.e., Microsoft Forms). Other forces have developed PowerApps (see Operation Rasure, Thames Valley Police). The minimum data requirement is the unique location visited (preferably geographic coordinates footprint), date and time patrol commenced from and to.
Tracking compliance. This can include monitoring whether officers are remaining within the geographic constraints of the micro-space, the frequency of visits (are the treatments assigned visited as many times as they were meant to be), the duration of visits (was the optimum 15 minutes achieved, how many minutes did the patrol last), and timing of visits (are they predictable or intermittent, are they taking place during hours when offence volumes are typically higher). This information is required to assess and evaluate the effectiveness of the project.
7. Evaluate effectiveness
7. Evaluate effectiveness
Evaluating effectiveness, in crime reduction terms, is going to depend on the study design. A really useful place to begin when thinking about evaluations (but from the onset rather than when you arrive at this point in a project) is the 'Assessing Responses to Problems: Did it work?' guide available on the popcenter website. Other available guides on designing and evaluating policing projects can be found at the College of Policing, the Policing Evaluation Toolkit. There is also a Microsoft Excel resource called the ABC Spreadsheet which is designed to measure before and after crime hotspot interventions, and displacement.
Recently available statistical evaluation methods for hotspot policing studies can be found in the following study examples:
Effects of One-a-Day Foot Patrols on Hotspots of Serious Violence https://link.springer.com/article/10.1007/s41887-021-00067-2
Fifteen Minutes per Day Keeps the Violence Away https://link.springer.com/article/10.1007/s41887-021-00066-3
Predictable Policing: Measuring the Crime Control Benefits of Hotspots Policing at Bus Stops https://link.springer.com/article/10.1007/s10940-016-9312-y
"Soft" policing at hot spots, Peterborough UK https://link.springer.com/article/10.1007/s11292-016-9260-4
Sweet spots of residual deterrence https://osf.io/preprints/socarxiv/kwf98/
References
References
Ashby, M.P.J. (2017). Comparing Methods for Measuring Crime Harm/Severity. Policing: A Journal of Policy and Practice, 12(4), pp.439–454. https://discovery.ucl.ac.uk/id/eprint/10076721/ [Accessed 6.11.2021]
Bachman, R and Schutt, R.K. (2020). Fundamentals of research in criminology and criminal justice. Thousand Oaks, California: Sage Publications, Inc.
Basford, L., Sims, C., Agar, I., Harinam, V. and Strang, H. (2021). Effects of One-a-Day Foot Patrols on Hot Spots of Serious Violence and Crime Harm: a Randomised Crossover Trial. Cambridge Journal of Evidence-Based Policing. https://link.springer.com/article/10.1007/s41887-021-00067-2 [Accessed 6.11.2021]
Bland, M., Leggetter, M., Cestaro, D. and Sebire, J. (2021). Fifteen Minutes per Day Keeps the Violence Away: a Crossover Randomised Controlled Trial on the Impact of Foot Patrols on Serious Violence in Large Hot Spot Areas. Cambridge Journal of Evidence-Based Policing. https://link.springer.com/article/10.1007/s41887-021-00066-3 [Accessed 6.11.2021]
Braga, A.A., Turchan, B., Papachristos, A.V. and Hureau, D.M. (2019). Hot spots policing of small geographic areas effects on crime. Campbell Systematic Reviews, 15(3). https://onlinelibrary.wiley.com/doi/full/10.1002/cl2.1046 [Accessed 6.11.2021]
Chainey, S. (2020). Understanding crime : analyzing the geography of crime. Redlands, California: Esri Press.
Chainey, S.P., Matias, J.A.S., Nunes Junior, F.C.F., Coelho da Silva, T.L., de Macêdo, J.A.F., Magalhães, R.P., de Queiroz Neto, J.F. and Silva, W.C.P. (2021). Improving the Creation of Hot Spot Policing Patrol Routes: Comparing Cognitive Heuristic Performance to an Automated Spatial Computation Approach. ISPRS International Journal of Geo-Information, 10(8), p.560. https://www.mdpi.com/2220-9964/10/8/560 [Accessed 6.11.2021]
College of Policing (2022) Serious Violence Hotspots Policing Guide. https://www.college.police.uk/guidance/serious-violence-hot-spots-policing-guide [Accessed 23.7.2023]
Eck, J. (2017). Assessing responses to problems: Did it work? An Introduction for police problem-solvers 2nd Edition. Center for Problem-Oriented Policing. https://popcenter.asu.edu/sites/default/files/assessing_responses_to_problems_final.pdf [Accessed 8.11.2021]
Hutt, O.K., Bowers, K. and Johnson, S.D. (2021). The effect of GPS refresh rate on measuring police patrol in micro-places. CrimRxiv. https://crimesciencejournal.biomedcentral.com/articles/10.1186/s40163-021-00140-1 [Accessed 6.11.2021]
Kime, S. & Wheller, L. (2018). The Policing Evaluation Toolkit. College of Policing, NPCC, AOPCC. [Accessed 8.11.2021] https://whatworks.college.police.uk/Support/Documents/The_Policing_Evaluation_Toolkit.pdf
Koper, C.S. (1995). Just enough police presence: Reducing crime and disorderly behavior by optimizing patrol time in crime hot spots. Justice Quarterly, 12(4), pp.649–672.
Linton, B. and Ariel, B. (2020). Random Assignment with a Smile: How to Love “TheRandomiser.” Cambridge Journal of Evidence-Based Policing, 4(3-4), pp.233–237. https://www.repository.cam.ac.uk/handle/1810/315597 [Accessed 6.11.2021]
Mitchell, R.J. (2019). Hot spots policing made easy. In: Evidence Based Policing An Introduction. Bristol: Policy Press.
Ratcliffe, Jerry H. (2019) ABC spreadsheet calculator (version 1.4) [Computer Software], reducingcrime.com.
Ratcliffe, J.H. and Sorg, E.T. (2017). Foot patrol : rethinking the cornerstone of policing. Cham: Springer.
Ratcliffe, J.H., Taniguchi, T., Groff, E.R. and Wood, J.D. (2011). The Philadelphi Foot Patrol Experiment: A Randomized Control Trial of Police Patrol Effectiveness in Violent Crime Hotspots. Criminology, 49(3), pp.795–831. https://www.jratcliffe.net/publications [Accessed 6.11.2021]
Rossy, Q. (2019) Visualist: a spatial analysis plugin for crime analysts. Ecole des sciences criminelles, Lausanne. https://plugins.qgis.org/plugins/visualist/ [Accessed 6.11.2021]
Sherman, L.W. and Weisburd, D. (1995). General deterrent effects of police patrol in crime “hot spots”: A randomized, controlled trial. Justice Quarterly, 12(4), pp.625–648. https://www.researchgate.net/publication/232860936_General_deterrent_effects_of_police_patrol_in_crime_HOT_SPOTS_A_randomized_controlled_trial [Accessed 6.11.2021]
Telep, C. and Hibdon, J. (n.d.). Problem-Oriented Guides for Police Problem-Solving Tool Series No. 14 Understanding and Responding to Crime and Disorder Hot Spots. [online] Available at: https://www.smart-policing.com/sites/default/files/2020-04/SPI-2019-HotSpots-v1.pdf [Accessed 6.11.2021]
Tompson, L., Johnson, S., Ashby, M., Perkins, C. and Edwards, P. (2014). UK open source crime data: accuracy and possibilities for research. Cartography and Geographic Information Science, 42(2), pp.97–111. http://irep.ntu.ac.uk/id/eprint/25672/1/220996_PubSub2745_Ashby.pdf [Accessed 6.11.2021]
Weinborn, C., Ariel, B., Sherman, L.W. and O’ Dwyer, E. (2017). Hotspots vs. harmspots: Shifting the focus from counts to harm in the criminology of place. Applied Geography, 86, pp.226–244. https://daneshyari.com/article/preview/6458335.pdf [Accessed 6.11.2021]
Weisburd, D., Petrosino, A. and Mason, G. (1993). Design Sensitivity in Criminal Justice Experiments. Crime and Justice, 17, pp.337–379.
Weisburd, D. (2015). The Law of Crime Concentration and the criminology of place. Criminology, 53(2), 133–157. https://onlinelibrary.wiley.com/doi/abs/10.1111/1745-9125.12070
Resources
Resources
Cambridge Crime Harm Index https://www.crim.cam.ac.uk/research/thecambridgecrimeharmindex
Office for National Statistics, Crime Severity Score https://www.ons.gov.uk/peoplepopulationandcommunity/crimeandjustice/datasets/crimeseverityscoreexperimentalstatistics
Open Street Map data extracts http://download.geofabrik.de/
Optimal Design http://hlmsoft.net/od/ (software for statistical analysis, statistical power test)
Philadelphia Foot Patrol Experiment https://www.jratcliffe.net/phila-foot-patrol-experiment
The Randomiser https://www.therandomiser.co.uk/admin/
UK Police Force data https://data.police.uk/
Page updated
Google Sites
Report abuse