Algorithm for the grouping of contiguous geographical zones

In geographical epidemiology, the correlation between two variables is often studied, using as basic units those zones for which these variables are available. However, because of their geographical contiguity, it could be misleading to consider each of these zones as independent statistical units. Moreover, spatial distributions (either of the factor or of the disease) are often distributed in relatively simple patterns; these patterns may induce strong correlations. Thus, the general pattern of the spatial distribution of the variables should be investigated before any attempt to infer epidemiological interpretation of a relationship. This paper presents an algorithm to construct contiguous groupings which are similar enough with regard to one of the two variables considered and to thus divide the original set of geographical areas into contrasting zones. The algorithm is easy to compute and allows mapping of the geographical distribution of the variable studied. It also lessens the risk of duplication of basic information by reducing the autocorrelation of one of the variables.