Vancouver break-and-enter hotspots

Discipline: Spatial statistics
Tools: ArcGIS
Methods: geocoding, Getis-Ord Gi* hot spot analysis, Global Moran's I, z-score and p-value reporting
Data: Vancouver Police Department; Vancouver Open Data

Notes on method

Hot spot maps are easy to produce and easy to over-read. A Gi* surface will always render something, and a reader who sees red assumes a finding.

So the global statistic is on the plate. Global Moran’s I returns a z-score of 39.28 against an expected index of roughly -0.0014, with a p-value that rounds to zero. That tells a reader the clustering is real before they start interpreting where it is, which is the order those two questions should be answered in.

Three confidence bands are shown rather than one. Collapsing 90, 95 and 99 percent into a single “hot spot” category would discard exactly the information that distinguishes a cluster worth acting on from one worth watching.

Hot spot map of Vancouver break-and-enter incidents with clusters graded by 90, 95 and 99 percent confidence, and a Global Moran's I summary table reporting a z-score of 39.3 and a p-value of zero. — Gi* clusters at three confidence levels. The Moran's I summary sits on the plate: z = 39.28, p = 0.000, so the clustering is not an artefact of the binning.Base data: Vancouver Police Department, Vancouver Open DataOpen larger (2432px)

Neighbourhood-level view restricted to high-significance break-and-enter clusters, with commercial building context described in the margin. — The same analysis aggregated to neighbourhoods, filtered to high-significance clusters only.Open larger (1397px)