Statistical Analysis
Four charts summarising the core findings: the top diffusion hubs by neighbourhood conversions, the diffusion rate compared to baseline, monthly crime volume across the study period, and the breakdown of crime types.
An extension of the Bexley hotspot project — using spatial network analysis to measure whether crime hotspots spread to neighbouring areas in subsequent months.
The main Bexley hotspot project uses a Random Forest model that treats every LSOA (Lower Super Output Area) as an independent observation — it has no concept of geography between areas. This extension challenges that assumption by asking: when an LSOA becomes a crime hotspot, do its geographic neighbours become hotspots in the following months?
This phenomenon is known in criminology as crime diffusion or near-repeat victimisation, and it has real implications for how police and community safety teams should allocate resources.
Each circle represents one LSOA. Circle size reflects number of network neighbours (degree); colour reflects total crime count — darker red means more crimes. Grey lines connect LSOAs within 1.2 km of each other. Click any node for details.
This map highlights which LSOAs act as diffusion hubs — areas that were hotspots and whose neighbours became hotspots the following month. Larger, darker circles indicate stronger hub behaviour. Grey circles had no recorded diffusion events.
Four charts summarising the core findings: the top diffusion hubs by neighbourhood conversions, the diffusion rate compared to baseline, monthly crime volume across the study period, and the breakdown of crime types.
When an LSOA became a hotspot, its neighbours had a 30.2% probability of becoming hotspots the following month — compared to a baseline rate of 25.8% for all LSOA-months. This 1.17× lift is statistically meaningful and consistent with near-repeat victimisation theory in criminology.
The spatial network contains 834 edges across 149 LSOAs, with an average of 11.2 neighbours per area. This high connectivity means hotspot pressure can ripple across large portions of the borough quickly, particularly through central hub areas.
Three LSOAs stand out as persistent diffusion hubs. Bexley 015A converted 145 neighbouring LSOA-months into hotspots — the highest in the borough — despite not having the highest absolute crime count. This suggests diffusion hubs are structurally central, not just high-crime.
The Random Forest model in the main project identifies individual hotspots in isolation. This network extension shows that proactive policing in hub LSOAs could suppress diffusion cascades — potentially reducing hotspot formation across multiple neighbouring areas simultaneously.
The analysis follows a six-step pipeline from raw crime data to diffusion measurement.
59,748 street-level incidents loaded from the Metropolitan Police dataset, covering Bexley Borough across 2023–2026.
LSOA-months aggregated by crime count. Any month with count ≥ 75th percentile (14 incidents) labelled as a hotspot.
Each LSOA represented by the mean latitude/longitude of its recorded incidents — a close approximation of its true geographic centroid.
igraph network built using Haversine distances. LSOAs within 1.2 km connected by an edge weighted by distance.
For each hotspot LSOA-month, all neighbours checked for hotspot status in the following month. Diffusion rate and lift calculated.
Each LSOA scored by total neighbour conversions while it was a hotspot — identifying areas most likely to seed crime in surrounding neighbourhoods.