An information statistical approach to the modifiable areal unit problem in incidence rate maps
The modifiable areal unit problem becomes apparent when incidence rates are mapped on the basis of areal units. Although small units with high spatial accuracy can present unreliable rates, large spatial units may remove relevant geographical variation. Regarding mapping as a kind of statistical modelling, this author proposes a new methodology to select appropriate areal units using the Akaike information criterion and two search methods for an informative geographical aggregation in map construction. The optimal zoning of similarity is suitable for finding spatial anomalies but presents a biased overall pattern. An alternative approach is to cluster areal units according to explanatory variables: this shows clear spatial patterns of elderly men's mortality matching the ecological structure in the Tokyo metropolitan area.