"Geography is destiny. Where crime happens tells us as much about urban structure as it does about criminal behavior."
To understand crime in Boston, we must first understand Boston itself—a city of distinct neighborhoods, each with its own character, history, and relationship to its neighbors. We used iGraph network analysis to model these relationships mathematically.
The Network Model
We constructed a graph where each police district becomes a node, and edges connect geographically adjacent districts. This seemingly simple model reveals profound patterns about how crime flows through the city.
Centrality Analysis
Network centrality measures identify which districts occupy strategic positions in the urban fabric. High centrality means more connections, more traffic, and often more crime spillover.
| District | Degree Centrality | Betweenness | PageRank |
|---|---|---|---|
| B2 (Roxbury) | 0.45 | 0.312 | 0.124 |
| C11 (Dorchester) | 0.36 | 0.198 | 0.108 |
| D4 (South End) | 0.36 | 0.167 | 0.096 |
| A1 (Downtown) | 0.27 | 0.089 | 0.082 |
| E13 (Jamaica Plain) | 0.27 | 0.071 | 0.079 |
District B2 (Roxbury) has the highest betweenness centrality—meaning it sits on the shortest paths between many other districts. This strategic position may explain its consistently high crime rates: it's a transit zone.
Community Detection
Using the Louvain algorithm for community detection, we identified natural clusters of districts that share more connections with each other than with outsiders. These communities often reflect historical neighborhood boundaries and socioeconomic patterns.
# Community detection using Louvain method
communities <- cluster_louvain(district_graph)
# Result: 3 distinct communities identified
# Modularity score: 0.412 (significant structure)Three distinct communities emerged from the analysis:
- Northern Cluster: A1, A7, A15 (Downtown, East Boston, Charlestown)
- Central Cluster: B2, B3, C6, C11, D4 (Roxbury, Mattapan, South Boston, Dorchester, South End)
- Western Cluster: D14, E5, E13, E18 (Brighton, West Roxbury, Jamaica Plain, Hyde Park)
Crime Flow Patterns
Network analysis also reveals how criminal activity might "flow" between adjacent districts. Districts with high connectivity tend to see spillover effects—when one district increases enforcement, crime may shift to neighbors.
Heavy policing in one area often pushes crime to adjacent districts rather than eliminating it. Network-aware patrol strategies can account for this by coordinating across district boundaries.
Geographic Coordinates as Features
Our machine learning models (explored in Chapter 4) found that latitude and longitude are among the most powerful predictive features. This confirms what the network analysis suggests: where matters enormously.
Together, geographic coordinates account for nearly 50% of our model's predictive power. No other feature category comes close. Boston's crime is fundamentally a spatial phenomenon.