I am carrying out non-hierarchical cluster (kRCluster) analysis on presence/absence data in PRIMERv7. Is it necessary to do both Bray Curtis & Jaccard similarities? I have seen both carried out but presume that previous studies had data that they subsequently converted to presence/absence. Any insight would be appreciated.
https://www.researchgate.net/post/Does-it-make-any-sense-to-use-both-Jaccard-index-and-Bray-Curtis-coefficient-to-compare-similarity-matrices-of-two-datasets
btw, these are distance measures, so there is no need for 'ground truth' as mentioned above?
Is it necessary to do both Bray Curtis & Jaccard similarities? Absolutely not! Jaccard is a presence/absence measure sjk = 100[(a/(a+b+c)] where a is number taxa in j and k (i.e. 1, 1), b is the number in j but not k (i.e. 1, 0) and c the number in k but not in j (i.e. 0, 1). Like all good biological measures it ignores double absences (0, 0). Bray-Curtis on presence/absence data is also called the Sorensen or Dice coefficient, and the formula for that is very similar: 100[2a/(2a+b+c)]. The 2 measures are monotonically related, and if you are doing any sort of rank-based analyses (e.g. ANOSIM or nMDS) both will give the same results.