Dear R-help! I ask you to help me with my problems with using R.
First, I ask you to forgive my bad English!
I try to use R in my study.
Subject of my work is comparative study of flora of lakes in different
regions of Russia. I have done floristical descriptions of 152 lakes (I
think it's enough) and have tabulated it. As data I have a table, such has
152 rows (lakes) 290 variables (species of plants). Thus, there is frequency
of species of plants in every cell of this table. I wanted to search some
groups between these lakes. I have done this searching with cluster analysis
(cutree(hclust(dist(DATA,"manh"),"ward"),4)). Then I apply
principal
component analysis:
=============##loading data (DATA)
##loading the list of groups (GROUPS)
d.prc<-princomp(DATA)
palette(rainbow(length(unique(GROUPS))))
plot(d.prc$scores,type="n",main="Principal Component
Analysis",xlab="Different groups of lakes have different
color-labeling")
text(d.prc$scores,labels=GROUPS,col=GROUPS,cex=.6)
=============I have received attached plot as a result. Both I and my supervisor
of
studies are agree with clustering. You can see 4 groups in this figure.
Write to me please your opinion, if these groups are authentic. Could you
present any criteria of existence of distinguishable groups and any criteria
of allocation of these groups?
Can I use PCA as instrument to test visually clustering?
I heard some opinion: "PCA is for visual allocating of the groups and then
you can search hypothetical parameter to explain groupping. But you cannot
use PCA for looking for the loadings (influence) of variables on the
groupping". Is it true?
Could you advise to me some other methods to analyze.
-----------------------------------
Best wishes,
Altshuler Eugenij P.
mailto:lapotcarex at mtu-net.ru
