Hi to all!
Recently, I've been trying to define functional groups of species by traits. One of the most common measures is the total dendrogram branch length to measure functional diversity (proposed by Petchey & Gaston, 2002; 2006). So, I made a cluster analysis, but I want to validated using some type of statistical approach.
I believe that I have two options... using Bootstrapping to determine p-values for each node... (available in pvclust R package) or the SIMPROF routine (using PRIMER).
I would like to know your opinions and insights about this methods!
the best
Using the term 'validation' here does not appear appropriate. Bootstrap, that you mention, will provide you a measure of confidence in your clustering results. See if Chapters 5 and 6 of these notes may possibly help to get some idea
https://www.researchgate.net/publication/268747239_Classification_and_Ordination_Tools_for_Biodiversity_Analysis
Deleted research item The research item mentioned here has been deleted
You can use the bootstrap to evaluate the probability of support of each branch of a dendograam which is a popular approach. Bradley Efron has an article describing the approach. See Efron et all, Bootstrap confidence levels for phylogenetic trees', Proc. Natl. Acad. Sci. USA (93)1996.
Dear Fabio,
Bootstrap techniques are used to estimate confidence intervals. They do not simulate a null hypothesis and therefore cannot estimate a p-value (at least, not honestely). Still, they have been used abusevely for that purpose.
Dear Fabio,
The bootstrap is used to estimate the sampling distribution of any statistic and I agree with Vasco that they do not simulate null hypotheses. They can estimate the proportion of bootstrapped samples that agree on clades of the original dendogram. These are not p-values. They are estimated proportions and referred to as bootstrap confidence values (for the clades of the trees). by Efron in his paper.
thanks to all for you answers, I'm understanding better this type of approach...
After all,, i would like to know what type of analysis do you suggest me if i want to be confident a about my functional diversity dendrogram????
If you want to estimate p-values, you could use permutation tests. That is the other type of MonteCarlo statistics, alternative to the one congregating bootstrap and jacknife. It simulates the null hypothesis randomly redistributing s.u among treatments (in an ANOVA design) or randomly redistributing each variable independently of each other (when testing significances of correlations and regressions). Now, if you have a multivariate data set, including both numerical and categorical variables, you must decide when you randomly resample, about what information do su loose their past identity and what do they preserve.
- Badges
- Science method