R help - Jul 2012 - Error in hclust?

Dear R users,

I have noted a difference in the merge distances given by hclust using
centroid method.

For the following data:

x<-c(1009.9,1012.5,1011.1,1011.8,1009.3,1010.6)

and using Euclidean distance, hclust using centroid method gives the
following results:
> x.dist<-dist(x)
> x.aah<-hclust(x.dist,method="centroid")
> x.aah$merge
     [,1] [,2]
[1,]   -3   -6
[2,]   -1   -5
[3,]   -2   -4
[4,]    1    2
[5,]    3    4
> x.aah$height
[1] 0.50000 0.60000 0.70000 0.97500 1.36875

A calculation by hand results same merges, but at different distances for
latter stages:

heights:
   0.5  => merging 3 and 6  => G1
   0.6  => merging 1 and 5  => G2
   0.7  => merging 2 and 4  => G3
   *1.25  => merging G1 and G2 => G4
   1.92  => merging G3 and G4*

It seems that hclust is not correctly computing the group centroids. Is it
correct?

Best regards,

Mateus

	[[alternative HTML version deleted]]

Mateus,

I see what you mean.
I can't figure it out.
I found a message from a few years ago that mentions the same problem.
http://tolstoy.newcastle.edu.au/R/e8/help/09/12/8165.html

This may be a bug.
Since hclust() is in the stats package, I am ccing the maintainer, 
R-core@r-project.org, on this e-mail.

Jean
> find("hclust")
[1] "package:stats"
> maintainer("stats")
[1] "R Core Team <R-core@r-project.org>"



Mateus Teixeira <mateus.teixeira@gmail.com> wrote on 07/04/2012 07:39:39 
AM:
> Dear R users,
> 
> I have noted a difference in the merge distances given by hclust using
> centroid method.
> 
> For the following data:
> 
> x<-c(1009.9,1012.5,1011.1,1011.8,1009.3,1010.6)
> 
> and using Euclidean distance, hclust using centroid method gives the
> following results:
> 
> > x.dist<-dist(x)
> > x.aah<-hclust(x.dist,method="centroid")
> > x.aah$merge
>      [,1] [,2]
> [1,]   -3   -6
> [2,]   -1   -5
> [3,]   -2   -4
> [4,]    1    2
> [5,]    3    4
> > x.aah$height
> [1] 0.50000 0.60000 0.70000 0.97500 1.36875
> 
> A calculation by hand results same merges, but at different distances 
for
> latter stages:
> 
> heights:
>    0.5  => merging 3 and 6  => G1
>    0.6  => merging 1 and 5  => G2
>    0.7  => merging 2 and 4  => G3
>    *1.25  => merging G1 and G2 => G4
>    1.92  => merging G3 and G4*
> 
> It seems that hclust is not correctly computing the group centroids. Is 
it
> correct?
> 
> Best regards,
> 
> Mateus
	[[alternative HTML version deleted]]