Statistics Glossary → Spatial autocorrelation (Moran's I, Geary's C)
Spatial autocorrelation (Moran's I, Geary's C)concept
The tendency of a variable to take similar (or dissimilar) values at nearby locations; its common measures are Moran's I and Geary's C.
It is the measured form of Tobler's principle that 'near things are more related than distant things'. Moran's I multiplies each location's deviation from the mean by the deviations of its neighbours, as defined by a spatial weights matrix W, and sums them; its structure resembles a correlation coefficient. Its expected value under randomness is -1/(n - 1), very close to zero; values clearly above it indicate positive autocorrelation (clustering of similar values), and values below it negative autocorrelation (a checkerboard pattern). Geary's C is based on the squared differences between neighbouring values; 1 indicates randomness, values below 1 positive autocorrelation, and it is more sensitive to local differences.
The result depends strongly on how neighbourhood is defined (shared borders, distance bands, k nearest neighbours), and different Ws can give different results. Global Moran's I gives a single summary for the whole area; local indicators (LISA, Getis-Ord Gi*) show where the clusters are. For rate data (e.g. disease rates), rates in small-population areas fluctuate a lot, so an empirical Bayes adjusted Moran's I can be used. If regression residuals are spatially autocorrelated, the observations are not independent and the usual standard errors cannot be trusted; spatial regression models are needed.
An everyday example
Average house prices are mapped by neighbourhood in a city. Expensive neighbourhoods line up along the coast and cheap ones around the industrial zone. Moran's I = 0.58 with p < 0.001: prices are not randomly distributed, and neighbourhoods with similar prices cluster together.
In MerQur
Available in the Moran's I analysis as global and local Moran and through the 'Additional statistic' option (Geary's C, Bivariate Moran, EB rate Moran). The 'ISA' check box in Getis-Ord Gi* computes the global Moran's I z value at several distance bands to help choose the band; the KDE density map and hexbin map address the same question of spatial pattern.
Analyses where this concept appears
KDE Density MapHexbin MapMoran's I (Spatial Autocorrelation)Getis-Ord Gi* Hotspot
Related concepts
Local indicators of spatial association (LISA) and Gi*Spatial weights matrix (W)