Abstract
The aim of this paper is to analyse different alternative implementations for a problem defined as "point allocation inside polygons" for Geographically Weighted Regression (GWR). The problem involves situations where the precise location of each observation is not known - just its district, municipality or region, i.e. a polygon geographical location. However, associated data were available that could potentially allows point placement of observations. These analyses were applied in a Income predicting model based on electricity consumption from a survey for a power distribution company in Sao Paulo, Brazil. Completely spatially random allocation and allocations based on spatial distributions of population (universe) and of the independent variable (electricity consumption) were utilized. Results showing the coefficients of determination (R2) suggest that a more realistic measure of the relationship between these two constructs could be evaluated.