Hello all, I am conducting a Bray-curtis / UPGMA analysis in R, based in a binomial dataset (presence and absence of species in specific areas).
I want to cluster the areas according to their similiarity in terms of species richness, but am unsure of what agglomeration method I should use in the hclust function.
I think that using Average or Median, with binary data, might not be the most appropriate method for this function. I have read that you can employ Ward or Complete.
Can anyone give me a nudge in the right direction please?
Thank you in advance.
Ali Mohammed Yassen Al Hashemi , an Unweighted Pair Group Method with Arithmetic Mean, as a clustering analysis method.
Although the clustering methods depend on the proximities, I would recommend you to consider both of them separately at the beginning.
On the one hand, we have the proximities. Given your dataset, I would recommend you to use a proximity measure that is common to find in other works with binomial data, for example Jaccard, Lance and Williams, Yule's Y or even the Euclid distance. In the end, the proximity is measuring how "close/far away" are your observations/variables with respect to one another in your dataset. Intuitively, you are extracting the information of similarity/dissimilarity among your data, and therefore, your proximity/dissimilarity measure should try to take into account what is in your data. For example, with high-correlated interval data, I will recommend using Mahalanobis and, I will not trust any Euclid distance. But, in your case, I recommend you to use several proximities/dissimilarities and compare the results among them.
On the other hand, with respect to the clustering method. UPGMA is ok. In the end, UPGMA is taking into account your proximities/dissimilarities measures independently of how the original data was. Nonetheless, I recommend trying different pairs of proximities/dissimilarities - clustering method. Given that you are working with a Hierarchical one, I recommend you these pairs (among many others):
Jaccard - Ward clustering method
Y's Yule - Simply linkage
Y's Yule - Complete Linkage.
These two last cases work in opposite ways, therefore may provide you with useful insights. I hope I was helpful. Clustering is quite tricky.
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking