I am helping someone analyse data on plant species. They divided a town into equal-sized squares and collected data on the weed species growing in each square. I used cluster analysis (Ward.D2) to group the squares according to their species composition. The results look sensible. There is data available about the ecological characteristics of most of the species (Ellenberg values etc). Can I now do a second analysis, applying the Ellenberg values to the species in each cluster and using ANOVA to see which Ellenberg values (e.g. nitrogen, pH) are most important in separating the squares? Or are there other ways of doing this?
Thank you for all the comments. The clustering was done on species presence/absence (indicator variables). Some species occur almost everywhere and others only occur once or twice in the whole data set. The auxiliary variables are Ellenberg indicator values (EIVs) for these species. In response to Roger V’s comment, I used pvclust (10000 iterations) to test the strength of the clustering. The 2 main clusters A and B were significant at 92%AU. Cluster A segmented into 2 subclusters and B into 3 subclusters which map well onto what is known in broad terms about the ecological and geographical characteristics of the location – urban character etc., the AUs of these subclusters vary a lot and are very poor for two species-poor subclusters. I have then calculated unweighted Ellenberg values for the species at both the main and subcluster levels. ANOVA of Ellenbergs at the A and B level are all p<0.001 and also mostly good p’s at the sub-cluster level with the exception of the 2 subclusters with poor AUs which are the most heavily urban. My reasoning (which may well be wrong!) is that the clustering has extracted sites which are similar in their species composition and then calculating mean EIV’s for the species in each cluster gives me information about the environmental factors which are most important in determining these clusters – eg nitrogen, pH. I have been reading Zeleny and Schaffers 2012 (J Veg Science) about circularity of reasoning with EIV’s, and I am not sure whether I am falling into a trap or not. Further, I decided not to weight Ellenberg values by the number of occurrences of a species in a particular square as the most abundant species tend to be high nitrogen and high light and so these attributes tend to dominate the data set. (Although this is important in many respects).
