Additional Comments on Weighting Values in Correlation Analysis of Areal Data

The magnitudes of simple regression and correlation parameters may be merely a function of the sizes and numbers of areal units into which a study area is divided. The areal weighting of data, a suggested solution to this problem, is effective only: 1) when the values for areal units which are aggregated are equal and 2) when any changes in summary values are strictly proportional. In a stochastic approach to the problem differences in parameters for the same study which is differently subdivided are hypothesized as chance occurrences. Four sets of examples are presented. When each of four study areas are changed by dividing them into different numbers and sizes of areal units, the resulting differences in regression and correlation parameters are merely chance occurrences.