--- title: R: Cocopan --- What does COCOPAN do ? This program carries out a one-way analysis of variance for spatially autocorrelated quantitative data when the classification criterion is a subdivision of the study area into nonoverlapping regions -- for instance countries, counties, language areas, geomorphological subdivisions of the study area, and so on, as found in numerous problems where the data points can be plotted on a map. The method has been described by Legendre, Oden, Sokal, Vaudor and Kim (1990). The acronym COCOPAN stands for Contiguity-constrained permutational ANOVA. The principle of this permutation test is to keep the localities fixed, each preserving its values for the variables under study, so as to preserve the autocorrelation structure. The classification criterion, which consists of the division of the map into subregions, is permuted instead, with the following constraints: each pseudo-region must contain the same number of localities as the original region that it represents; each pseudo-region must remain connected, that is, it must form a continuous surface on the pseudo-map; finally, the pseudo-regions must occupy exactly the whole of the original map, without omission or excess. The program contains two algorithms to resolve the computation problem; the ring algorithm, created by Alain Vaudor, and the random tree algorithm, developed by Junhyong Kim. Several variables may be analyzed in a single run. The statistic subjected to permutation testing is the sum, for all groups, of the within-group sums of squares (SSQ). After each permutation, the SSQ statistic is recomputed for that pseudo-map. Finally, the SSQ value for the true map is compared to the distribution of the SSQ values obtained for the pseudo-maps. So, this is a one-tailed test and the critical area is the lower tail of the distribution.
Last updated on Saturday, March 30, 2013 by Philippe Casgrain