Most commonly this occurs when the underlyingspatial process creates regions of attribute clustering that are much larger thanthe units of observation chosen by (or available to) the analyst. Figure 1 revealsthat the areas of high and low poverty generally are considerably moreextensive than is the particular lens (counties) through which we are viewingthe process. When units of analysis are much smaller than the regions of highor low attribute values, spatial autocorrelation in the observations is inevitable.As with substantive autocorrelation, nuisance autocorrelation must somehowbe recognized and eventually brought into the formal analysis. Anselin (1988,p. 15) differentiates among these types of autocorrelation using the terms‘‘apparent contagion’’ (spatial heterogeneity) and ‘‘real contagion’’ (spatialdependence).