Searching for the elusive aggregation effect: evidence from statistical simulations
The past few years have seen a resurging interest in the modifiable areal unit problem, or aggregation effects. The new evidence, however, both supports and conflicts with previous work. This paper represents the first stage in a series of numerical experiments designed to explore the nature and extent of scale and zonation effects. Results from a series of carefully controlled statistical simulations are reported. It is concluded that there definitely are aggregation effects separate from effects that can be attributed to changing the definition of the spatial process. These effects, however, vary with the statistic calculated. Means and variances are resistant to aggregation effects, whereas regression coefficients and correlation statistics exhibit dramatic effects. In summary, the world of spatial analysis as it relates to the modifiable areal unit problem is not entirely well-behaved, but neither is it completely random and ill-defined.