The procedures adopted for analyzing patterns of spatial autocorrelation depend on the type of data available. There is considerable difference between:

a)      a set of 100 values obtained for a 10x10 grid of (100mx100m) squares which covers a 1000mx1000m region;

b)      a set of values obtained for 100 contiguous but arbitrarily shaped polygons which again cover the same region;

c)       a set of nominal values (classes) for 100 contiguous but arbitrarily shaped polygons which again cover the same region; and

d)      a set of data values obtained from 100 arbitrarily distributed sample points in our study region

Each case warrants a slightly different approach, but each utilizes the notion of proximity bands as a means of imposing some form of serial behavior to the data. The idea is then to examine the correlation between areas or points at given levels of separation, to obtain a similar measure to that used within time series analysis. Initially we examine the case of nominal-valued data (Section 5.5.2.1), which helps identify the key concepts involved and focuses our attention on a set of questions regarding such analysis that we examine in more detail in Sections 5.5.2.2 et seq.