Spatial Statistics of SADC Development Indicators

Development indicators across neighbouring countries are rarely independent: similar values tend to cluster geographically. A table of numbers hides that structure. The questions are whether the clustering is strong enough to be more than chance, and where the local hotspots and cold spots sit.

The interactive page pulls seven World Bank indicators for the 16 SADC countries and, for a chosen indicator and year, recomputes global Moran's I with its pseudo p-value and a per-country local Moran's I classification. A D3 parallel-coordinates plot and a MapLibre map share a selection, so brushing a band of values highlights the matching countries on the map and vice versa. Separately, a geostatistics study in R uses point-referenced data to fit variograms and interpolate a continuous surface with uncertainty.

• k-nearest-neighbour spatial weights over the 16 SADC countries
• Global Moran's I with permutation inference (9,999 permutations)
• Local Moran's I (LISA) classifying high-high, low-low and spatial-outlier countries
• Linked D3 parallel-coordinates plot and MapLibre GL map with shared brushing and selection
• Separate geostatistics study in R (gstat): empirical variograms and fitted spherical, exponential and Gaussian models
• Ordinary and universal kriging with leave-one-out cross-validation
• Conditional simulation to represent interpolation uncertainty

• For urban population share in 2022 the 16 SADC countries show positive global spatial autocorrelation (Moran's I = +0.425), highly unlikely under spatial randomness (pseudo p = 0.0001 over 9,999 permutations), so neighbouring countries are more alike than chance alone would predict.
• Local Moran's I resolves that global figure into specific high-high and low-low clusters, with a handful of countries flagged as spatial outliers against their neighbours.

• Coursework for a spatial-statistics module. The interactive page is hosted on AWS (CloudFront) with the basemap served through Amazon Location Service; the kriging study uses a separate point-referenced dataset.