Cluster: Avoiding distance "rescaling"?

[Prof1950]

My data is already in a distance metric that I would like to cluster directly, without doing a Euclidean "rescaling" on it. Is there any way to do this with Cluster in SPSS?

[Art Kendall]

  Which of the few dozen distance/similarity measures is used is only
relevant when you are reading in raw data.

Was your distance matrix produced by SPSS? just use the "/matrix in
'filespec'" option.

If you have the raw data it is likely to be easier to read it in and
just specify the distance/similarity measure you want to use.
However, you can also check <help> and format your input file the way
that SPSS formats matrix input/out put.  Sometimes it helps to
use CLUSTER or PROXIMITIES on any data and see how "/matrix out" formats
the file.



Art Kendall
Social ResearchConsultants

On 9/19/2010 6:58 PM, Prof1950 wrote:

> My data is already in a distance metric that I would like to cluster
> directly, without doing a Euclidean "rescaling" on it. Is there any way to
> do this with Cluster in SPSS?
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Cluster-Avoiding-distance-rescaling-tp2846046p2846046.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Prof1950]

Actually, my data is already in the distance metric that I want to use, so I don't want to transform it by ANY of the multiple options available. (Sorry if I wasn't clear that I was just using Euclidean as an example.) As far as I can tell, I have to rescale my data using one of SPSS's distance measures, no matter what I prefer.

This is an extreme "work around," but perhaps there is one of these measures where it would be relatively easy to recode my input as its own "reciprocal," so that when SPSS applies that particular rescaling, it will convert my data back to their original form??

[Art Kendall]

see interspersed comments.
I think I see how you thought this. The documentation says.

MEASURE Subcommand (CLUSTER command)

MEASURE specifies the distance or similarity measure used to cluster cases.

• If the MEASURE subcommand is omitted or included without specifications, squared Euclidean distances are used.

• Only one measure can be specified.

The documentation could be clarified to say something like.

MEASURE Subcommand (CLUSTER command)

MEASURE specifies the distance or similarity measure used to cluster cases.

• If the MEASURE subcommand is omitted or included without specifications, squared Euclidean distances are used. If matrix data is read in, the distance measure is not calculated.

• Only one measure can be specified.

Art

On 9/20/2010 12:02 PM, Prof1950 wrote:

Actually, my data is already in the distance metric that I want to use, so I
don't want to transform it by ANY of the multiple options available. (Sorry
if I wasn't clear that I was just using Euclidean as an example.) As far as
I can tell, I have to rescale my data using one of SPSS's distance measures,
no matter what I prefer.

You can verify that the distance matrix SPSS works with is the one you want with  "/print = distance" .  If your distances are for a large number of cases this can be voluminous, so you might not want to put it on paper. Once you have verified that you have what you want just run it again without the "/print"


If your data is already in the distance metric you want to use and it is already in SPSS matrix format do something like this
cluster /matrix in ('c:\project\mymatrix.sav') /method = ward. 

If your data is already in the distance metric you want to use and it is NOT already in SPSS matrix format edit it so that is in the format of an SPSS input/output matrix.
cluster /matrix in ('c:\project\mymatrix.sav') /method = ward.

I only suggest recalculating the distance measure
 IF you have the raw data
AND the distance measure you want is one that SPSS can calculate
AND you want to avoid the work of getting the matrix into SPSS matrix format.

Art Kendall
Social Research Consultants

This is an extreme "work around," but perhaps there is one of these measures
where it would be relatively easy to recode my input as its own
"reciprocal," so that when SPSS applies that particular rescaling, it will
convert my data back to their original form??
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Cluster-Avoiding-distance-rescaling-tp2846046p2846782.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD