Moving Towards Predictive Policing Professor Shane D Johnson UCL Department of Security and Crime Science shane.johnson@ucl.ac.uk Predicting future patterns • Questions we might ask – How many burglaries are expected in the next few days? – Bursty analysis (Johnson et al., 2012)? – Self-exciting Point Process? – What is likely to be stolen (Bowers & Johnson, 2012)? – Where will burglaries most likely next occur? – What is the relative risk within an area for (say) the next seven days – By day/night Overview • Spatial patterns (risk heterogeneity) – Patterns and predictors at the street segment level • Space-time clustering (event dependency) – What happens in the wake of an offense? – Point level analysis • Collaboration with West Midlands Police – Combining the approaches to analysis • Displacement? First things first: Spatial clustering of Burglary? Ordnance Survey © Crown Copyright. All Rights reserved Johnson, S.D., and Bowers, K.J. (2010). Permeability and Crime Risk: Are Cul-de-sacs Safer? Journal of Quantitative Criminology, 26, 113-138. Spatial Clustering at the Street Level? Highest risk segments: 5% of homes 40% of burglary Johnson, S.D. (2010). A Brief History of the Analysis of Crime Concentration. European Journal of Applied Mathematics, 21, 349-370. Crime Pattern Theory • Offender search patterns and personal activity space • Home to work to recreation – nodes and paths, and mental maps • Looking for opportunities • Paths people take and the nodes they inhabit explain their risks to victimisation Hypotheses H1 – the risk of burglary will be greater on Major roads and those intended to be most frequently used H2 – the risk of burglary will be highest on the most connected streets, particularly those connected to major roads H3 - the risk of burglary will be lower in cul-de-sacs and, in particular, in those that are non-linear Johnson, S.D., and Bowers, K.J. (2010). Permeability and Crime Risk: Are Cul-de-sacs Safer? Journal of Quantitative Criminology, 26, 113-138. Road classification • OS classification – – – – Major Minor Local Private • Manual classification (~11k street segments) – Linear – Non-linear cul-de-sacs (Sinuous) Cul-de-Sacs Mostly Linear Mostly Sinuous Ordnance Survey © Crown Copyright. All Rights reserved Aggregate Results by Segment Type Concentration at places: Repeat Victimization Is Victimization Risk Time-Stable? Timing of repeat victimization Johnson, S.D., Bowers, K.J., & Hirschfield (1997). New Insights into the Spatial and Temporal Distribution of Repeat Victimization. British Journal of Criminology, 37(2), 224-241. Explaining Repeat Victimisation Boost Account • Repeat victimisation is the work of a returning offender • Optimal foraging Theory (Johnson & Bowers, 2004) - maximising benefit, minimising risk and keeping search time to a minimum– repeat victimisation as an example of this – burglaries on the same street in short spaces of time would also be an example of this • Consider what happens in the wake of a burglary – To what extent is risk to non-victimised homes shaped by an initial event? Neighbour effects at the street level Bowers, K.J., and Johnson, S.D. (2005). Domestic burglary repeats and space-time clusters: the dimensions of risk. European Journal of Criminology, 2(1), 67-92. An analogy with disease Communicability • Communicability - inferred from closeness in space and time of manifestations of the disease in different people. ++ + ++ + ++ + + +++ + + area + burglaries Knox Analyses Previous analysis does not take account of patterns across streets The degree to which clustering occurs in Euclidian space can be measured using: - Monte Carlo simulation and Knox ratios (Knox, 1964; Johnson et al., 2007) D is ta n c e b e tw e e n e v e n ts in p a ir T im e b e tw e e n e v e n ts in p a ir 0 -1 0 0 m 1 0 1 -2 0 0 m 2 0 1 -3 0 0 m 7 d a ys 421 221 189 14 days 246 209 091 21 days 102 237 144 Johnson, S.D. et al. (2007). Space-time patterns of risk: A cross national assessment of residential burglary victimization. J Quant Criminol 23: 201-219. Patterns in detection data? For pairs of crimes: – Those that occur within 100m and 14 days of each other, 76% are cleared to the same offender – Those that occur within 100m and 112 days or more of each other, only 2% are cleared to the same offender Johnson, S.D., Summers, L., Pease, K. (2009). Offender as Forager? A Direct Test of the Boost Account of Victimization. Journal of Quantitative Criminology, 25,181-200. Near Repeats – Foraging What do offenders say? “If this area I didn’t get caught in, I earned enough money to see me through the day then I’d go back the following day to the same place. If I was in, say, that place and it came on top, and by it came on top I mean I was seen, I was confronted, I didn’t feel right, I’d move areas straight away …” (P02) “The police certainly see a pattern, don’t they, so even a week’s a bit too long. Basically two or three days is ideal, you just smash it and then move on … find somewhere else and then just repeat it, and then the next area …” (RC02) Summers, L., Johnson, S.D., & Rengert, G. (2010) The Use of Maps in Offender Interviewing. In W. Bernasco (Ed.) Offenders on Offending. Willan. Forecasting burglary risk Risk High Low Bowers, K.J., Johnson, S.D., and Pease, K. (2004). Prospective Hotspotting: The Future of Crime mapping? British Journal of Criminology, 44(5), 641-658. Computer Simulation Pitcher, A., & Johnson, S.D. (2011). Exploring Theories of Victimization Using a Mathematical Model of Burglary. Journal of Research in Crime and Delinquency, 48(1), 83-109. Forecast Accuracy Grid (50m X 50m cells) 2 2 2 10 3 n=1 R=1 Celli, j,t = å b Celli, j,t-a + åå b Bufferb,t-a + b HouseDensityi, j + å b Districti, j,n + å b Roadi, j,R a=1 a=1 b=1 KDE+ (Last 10) KDE+ (1st 10) Optimal (Last 10) 0.5 0.4 0.3 0.2 0.1 0.0 Fraction of Burglaries 0.6 Optimal (1st 10) 0.7 One BCU – Night (8pm to 8am) 0 2 4 6 % Cells 8 10 Model parameters may need updating: - Changes in offenders at liberty - Changes due to police strategy - Other factors But what’s the point of prediction, targeted policing will only displace the problem right? Summary and Combining the Approaches • Triangulation across methods • Burglary more likely at more connected segments – Analyses ignore patterns over time • Risk of crime temporarily elevated around victimized homes (predictable in space-time) – Topology of the street network ignored – Units of analysis “cells” not street segments • West Midlands Police and UCL Dept SCS Collaboration (Toby Davies) – Does risk diffuse along the street network in predictable ways? – Is risk more likely to be diffused along certain types of segment? – Other offence types – Randomized Controlled Trial