Dear researchers, I'm using a software called PAST to analyze my multivariate water quality data. For clustering the sampling stations using 14 water quality variables, I found that NJ with Euclidean similarity index method generates a good clustering pattern, but I'm a bit clueless in terms of interpreting the phylogram as overall. How can one tell the generated clusters by NJ are statistically significant? Are there any rule of thumb for bootstrap confidence values? and, is NJ really suitable for clustering water quality data?
I recommend the following. Check (e.g. using histograms or draftsmans plot) whether variables are skewed (e.g. some large outlying values). If they are, log transform them.
Normalise each variable to put all variables on a common (dimensionless) scale by subtracting the variable mean and dividing by the standard deviation.
Calculate Euclidean distances among samples.
There are various neighbor-joining options (nearest, furthest, group average). I recommend group average.
If you want to do testing of the 'significance' of structure in the resulting dendrogram, there is a method called Similarities Profiles analysis (SIMPROF). See Clarke KR, Somerfield PJ, Gorley RN (2008) Testing of null hypotheses in exploratory community analyses: similarity profiles and biota-environment linkage. J. Exp. Mar. Biol. Ecol. 366:56-69
Thank you Paul Somerfield . Pertaining to the mentioned transformation and normalization, it is necessary to do both procedures or the transformation per se is sufficient prior to data analysis?
The normalisation is more important, as it puts all the different variables, measured on differing scales, onto the same (dimensionless) scale. It should be done after the transformations, if they are necessary (and they usually are, especially for concentrations of things).
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation