Inferences based on spatial analysis of areal data depend greatly on the method used to quantify the degree of proximity between spatial units - regions. These proximity measures are normally organized in the form of weights matrices, which are used to obtain statistics that take into account neighbourhood relations between agents. In any scientific field where the focus is on human behaviour, areal datasets are immensely relevant since this is the most common form of data collection (normally as count data). The method or schema used to divide a continuous spatial surface into sets of discrete units influence inferences about geographical and social phenomena, mainly because these units are neither homogeneous nor regular. This article tests the effect of different geometrical data aggregation schemas on global spatial auto-correlation statistics. Two geographical variables are taken into account: scale (resolution) and form (regularity). This is achieved through the use of different aggregation levels and geometrical schemas. Five different datasets are used, all representing the distribution of resident population aggregated for two study areas, with the objective of consistently test the effect of different spatial aggregation schemas.