R help - Apr 2010 - Cluster analysis: dissimilar results between R and SPSS

Hello everyone!

My data is composed of 277 individuals measured on 8 binary variables
(1=yes, 2=no). 

I did two similar cluster analyses, one on SPSS 18.0 and one on R 2.9.2. The
objective is to have the means for each variable per retained cluster.

1) the R analysis ran as followed:
> call data
> dist=dist(data,method="euclidean")
> cluster=hclust(dist,method="ward")
> cluster
Call:
hclust(d = dist, method = "ward")

Cluster method   : ward
Distance         : euclidean
Number of objects: 277
> plot(cluster)
> rect.hclust(cluster, k=4, border="red")
> x=rect.hclust(cluster, k=4, border="red")
> sapply(x, function(i) colMeans(data[i,]))
> round(sapply(x, function(i) colMeans(data[i,])),2)
2) The SPSS analysis ran as follows:

Analysis --> Classify --> Hierarchical cluster analysis --> Cluster
methodWard''s method and Distance measure= Interval:  Squared Euclidean
distance.
After that, I computed the means of each variable for each cluster.

The problem is I have different results between the two analyses (different
clusters and means).

However, when I use the "Euclidean distance" (unsquared) in SPSS, I
have the
same results! 

I thought the R "euclidean" command meant the "usual square
distance between
the two vectors (2 norm)" as specified in the documentation, no the
unsquared distance. Did it not?

Thanks for the comment!

Jeffrey



	[[alternative HTML version deleted]]

Hi Jeoffrey,

How stable are the results in general ?
If you repeat the analysis in R several times, does it yield the same
results ?


Tal

----------------Contact
Details:-------------------------------------------------------
Contact me: Tal.Galili@gmail.com |  972-52-7275845
Read me: www.talgalili.com (Hebrew) | www.biostatistics.co.il (Hebrew) |
www.r-statistics.com (English)
----------------------------------------------------------------------------------------------




On Mon, Apr 26, 2010 at 3:37 PM, Jeoffrey Gaspard <
jeoffrey.gaspard@gmail.com> wrote:
> Hello everyone!
>
> My data is composed of 277 individuals measured on 8 binary variables
> (1=yes, 2=no).
>
> I did two similar cluster analyses, one on SPSS 18.0 and one on R 2.9.2.
> The
> objective is to have the means for each variable per retained cluster.
>
> 1) the R analysis ran as followed:
>
> > call data
> > dist=dist(data,method="euclidean")
> > cluster=hclust(dist,method="ward")
> > cluster
>
> Call:
> hclust(d = dist, method = "ward")
>
> Cluster method   : ward
> Distance         : euclidean
> Number of objects: 277
>
> > plot(cluster)
> > rect.hclust(cluster, k=4, border="red")
> > x=rect.hclust(cluster, k=4, border="red")
> > sapply(x, function(i) colMeans(data[i,]))
> > round(sapply(x, function(i) colMeans(data[i,])),2)
>
> 2) The SPSS analysis ran as follows:
>
> Analysis --> Classify --> Hierarchical cluster analysis -->
Cluster method> Ward's method and Distance measure= Interval:  Squared
Euclidean distance.
> After that, I computed the means of each variable for each cluster.
>
> The problem is I have different results between the two analyses (different
> clusters and means).
>
> However, when I use the "Euclidean distance" (unsquared) in SPSS,
I have
> the
> same results!
>
> I thought the R "euclidean" command meant the "usual square
distance
> between
> the two vectors (2 norm)" as specified in the documentation, no the
> unsquared distance. Did it not?
>
> Thanks for the comment!
>
> Jeffrey
>
>
>
>        [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
	[[alternative HTML version deleted]]

I'm not sure why you'd expect Euclidean distance and squared Euclidean
distance to
give the same results.

Euclidean distance is the square root of the sums of squared
differences for each variable, and that's exactly what dist() returns.

http://en.wikipedia.org/wiki/Euclidean_distance

On a map, it's the length of the hypoteneuse, and you can measure it
with a ruler
and get the same number. Euclidean distance has a specific geometric meaning.

Squared Euclidean distance is not the same thing, and not the standard
definition
you seem to be expecting. If that's what you want, then square the
output of dist()
before you perform the clustering.

Sarah

On Mon, Apr 26, 2010 at 8:37 AM, Jeoffrey Gaspard
<jeoffrey.gaspard at gmail.com> wrote:
> Hello everyone!
>
> My data is composed of 277 individuals measured on 8 binary variables
> (1=yes, 2=no).
>
> I did two similar cluster analyses, one on SPSS 18.0 and one on R 2.9.2.
The
> objective is to have the means for each variable per retained cluster.
>
> 1) the R analysis ran as followed:
>
>> call data
>> dist=dist(data,method="euclidean")
>> cluster=hclust(dist,method="ward")
>> cluster
>
> Call:
> hclust(d = dist, method = "ward")
>
> Cluster method ? : ward
> Distance ? ? ? ? : euclidean
> Number of objects: 277
>
>> plot(cluster)
>> rect.hclust(cluster, k=4, border="red")
>> x=rect.hclust(cluster, k=4, border="red")
>> sapply(x, function(i) colMeans(data[i,]))
>> round(sapply(x, function(i) colMeans(data[i,])),2)
>
> 2) The SPSS analysis ran as follows:
>
> Analysis --> Classify --> Hierarchical cluster analysis -->
Cluster method> Ward's method and Distance measure= Interval: ?Squared
Euclidean distance.
> After that, I computed the means of each variable for each cluster.
>
> The problem is I have different results between the two analyses (different
> clusters and means).
>
> However, when I use the "Euclidean distance" (unsquared) in SPSS,
I have the
> same results!
>
> I thought the R "euclidean" command meant the "usual square
distance between
> the two vectors (2 norm)" as specified in the documentation, no the
> unsquared distance. Did it not?
>
> Thanks for the comment!
>
> Jeffrey
>
>


-- 
Sarah Goslee
http://www.functionaldiversity.org