Spatial Poverty Mapping

Small-area estimation is usually the engine that produces the poverty numbers a map displays. It combines a detailed household survey, which measures consumption or poverty well but only for a sample, with a full census, which covers everyone but lacks income detail, to predict poverty rates for small administrative units. Spatial poverty mapping then visualises and analyses those estimates geographically, adds spatial covariates, and tests for clustering. The mapping is only as reliable as the small-area estimates underneath it, so their uncertainty should travel with the map.

Moran's I is a global measure of spatial autocorrelation — the degree to which nearby areas have similar values. It ranges roughly from -1 (a checkerboard of dissimilar neighbours) through 0 (random) to +1 (strong clustering of similar values). Poverty is typically positively autocorrelated, with poor areas next to poor areas. Measuring this matters both substantively, because clustering justifies geographic targeting, and statistically, because ignoring spatial dependence violates the independence assumptions of ordinary regression and biases inference.

Henderson, Storeygard, and Weil showed that the intensity of night-time lights captured by satellites is a useful proxy for economic activity, correlating with income and growth. In poverty mapping, night-lights and other remote-sensing layers serve as low-cost, frequently updated, globally available covariates that help explain the geography of poverty and predict well-being in places where household survey data are sparse, outdated, or unavailable, complementing rather than replacing direct poverty estimates.