I performed cluster analysis in R (using hclust under vegan package) to find relationships between habitats based on species. A dendrogram/tree was derived from Bray-Curtis sim. matrix. Cluster methods are Ward, Ward.D2, Single, Complete, Average etc.
However, when I perform an ANOVA with post-test, the significant differences between pairs of habitats do not represent the differences observed in the dendrogram, when using a variety of methods. E.g. ANOVA computed that A was different from C, but the cluster analysis showed that A is similar to C.
ANOVA was performed using species abundances and cluster analysis using Bray-Curtis which converts to binary/presence-absence.
Any ideas?
try a db-rda (capscale in Vegan in R). it's more or less similar to RDA but it tries to explain the variation of a similarity matrix based on Bray-Curtis distance
Hi Marie, thanks for the idea however, one needs a second predictor variable matrix to do RDA. I only have a community matrix: species x sites.
Are you using the same Type I error rate when you compare the results for the two methodologies? The ANOVA F test is very basic; it addresses the question "are all population means equal?" Which post-hoc test did you use? What does the cluster analysis test? Is there a p-value?
Hi. well i badly figured out what were your data. I think the problem here is that Bray-curtis distance is not suppoed to transformed your abundance data to presence-absence. and again, anova is not the right post-hoc test, try adonis :)
Prof. Raid, I used basic ANOVA with Dunn's post-hoc test, cluster analysis tested the relationship between habitats based on bird species presence/absence.
Attached are the cluster trees/dendrograms using different linkage methods. My ANOVA post-hoc test showed that Nam is significantly different from all other habitats (p
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA