“...criminal acts require convergence in space and time of likely offenders, suitable targets and the absence of capable guardians...” (Cohen and Felson, 1979) Discrete Choice Methods Graph Theoretic Methods Example Practical Outcomes Forthcoming Publications Example Theoretical Outcomes Analysing and predicting crime using discrete choice and graph theoretic methods Mr. Michael Frith ([email protected]), Prof. Shane Johnson and Dr. Hannah Fry Traditional methods for analysing this convergence either ana- lyse the target or guardianship but ignore the offender (regres- sion methods). Or they analyse the offender but ignore the target or guardianship (journey-to-crime methods). Discrete-choice methods can analyse these simultaneously. These methods work by modelling offence locations as a choice where the offender chose that location above all others. These models then relate that choice to the attributes of the person (the offender) or to the attributes of the alternatives available. Frith, M., Johnson, S.D., and Fry, H. An analysis of residential burglars and guardian- ship using graph theory and mixed logit. Submitted for peer review (Criminology). Frith, M., and Johnson, S.D. A comparative analysis of the offence location choices of different types of serious acquisitive crime offenders in York (UK). In preparation. Frith, M., Johnson, S.D., and Fry, H. A meta-analytic review of graph theoretic and space syntactic methods for estimating pedestrian and vehicular movement flows. In preparation. Traditional analyses also simplify the mechanics of this conver- gence and ignore how the street network determines navigation and movement. For example, how offenders find and reach tar- gets and where potential victims and guardians may accumulate. Here, graph theory can be used to analyse the configurational ef- fects of the built-environment. For example, existing metrics (e.g. betweenness) can be used to estimate aggregated movement flows and novel idiosyncratic metrics (in forthcoming paper) may estimate an offender’s movements and their ‘awareness space’. These methods and results will then be applied and evaluated for their practical uses in: Geographic Profiling Given a series of connected offences, can we use offence location preference data, and for example the new idiosyncratic graph theory metrics, to improve current methods for identifying the most probable area than an offender resides or otherwise frequents? Crime Forecasting Given offence location preference data (in- cluding on repeats and near-repeats), and the locations of offenders (or an assumption of their geographic distribution), can we im- prove on current methods for predicting of- fence locations and hotspots? Offender Profiling Given offence location data for a specific of- fender, we can estimate their specific prefer- ences. This can then be used, for example, for pre-emptive patrols or strategic housing (by probation) of ‘ex-offenders’ away from their preferred and tempting targets. Targets - When selecting targets, do offenders follow a hierarchical ‘hunting process’? If so, at what spatial units? (See A) - For non-stationary targets (e.g. robberies), can we use the victims’ estimated movements and their journey-to-crime to model and analyse their convergence with the offender? Guardianship - Can we accurately model movement flows using graph theory? And if so, can we do this for different times of the day using travel journal trip distributions? (See B) - Can we model the quality (and quantity) of passer-by guardianship using local-weighted graph theory metrics? Offenders - Can we accurately model an offender’s likely movements and their awareness space? - By allowing preference heterogeneity amongst offenders, how do their preferences vary? And can we use this to quantitatively build typologies of offenders? (See C) } } } Individual targets Area-level affluence Passer-by volume CCTV sites and lines of sight Proximity to offender’s home Awareness space Real world Morning Afternoon Evening ‘Target Area’ ‘Target Street’ ‘Actual Target’ A B C Distance Affluence Dislikes Likes Preference Heterogeneity Heterogeneity modelled into discrete classes Type A: Prefers closer targets Indifferent to affluence “Marauder” Overall Type A Type B Dislikes Likes Type B: Indifferent to distance Prefers affluent targets “Commuter” Interpret classes/typologies To City Centre Area A Area B Before Current Future (Actual) Predicted Traditional Method Using This Research Overall Type A Type B Hostel A (Not Preferred) Hostel B (Preferred) For pre-emptive patrols Most probable home location School Shops Bars School Shops Bars School Shops Bars