Rasterizing census geography: Definition and optimization of a regular grid

The results of first and second order analyses tend to be affected by the characteristics of the spatial units at which data are sampled. This paper discusses the definition of a regular grid superimposed on a set of given irregular census units, and the subsequent redistribution of the census variables to the newly defined grid cells. A statistical criterion guides the definition and optimization of the grid: through an objective function, the method aims at preserving the global spatial autocorrelation measured for a salient variable on the original census units. Several aspects of the grid positioning and population redistribution are critically discussed. Ultimately, the proposed method constitutes a valuable alternative to the spatial heterogeneity that affects many empirical spatial data.