K clustering preassigning some members in SPSS 14.0

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

K clustering preassigning some members in SPSS 14.0

OZGU GILL
Hi there,

  I have 110 cases and 10 variables which I want to assign into 12 clusters. I have run K clusters in SPSS 14.0 version. After checking resuults, some of the cases would be assigned to a different cluster. It will be more appropriate for some of the cases to be in a certain cluster. Is  there a way to assign these cases into some preselect clusters and then run the clustering.  Is there a way to assign weights to variables rather than the cases. In SPSS you can give weigts to cases there is no choice for variable that I have seen so far.  Thanks in advance.

  Ozgu


---------------------------------
Do you Yahoo!?
 Get on board. You're invited to try the new Yahoo! Mail Beta.
Reply | Threaded
Open this post in threaded view
|

Re: K clustering preassigning some members in SPSS 14.0

Spousta Jan
Hi Ozgu,

> Is  there a way to assign these cases into some preselect clusters and
then run the clustering?

With the given algorithm, it cannot be done directly, although K-means
itself can be generalized in this respect. In SPSS, I would replace the
values of the "known" cases with the values of their respective
centroids and so force that they will end in the same cluster. For
exmple if I knew that cases 3, 5 and 8 are from the same cluster
centered around (1, 7) in the feature space, I would write:

Do if (any(ident,3,5,8)).
- compute x = 1.
- compute y = 7.
End if.

And then run K-means.

> Is there a way to assign weights to variables rather than the cases?

Yes, it is. Before running K-means, multiply the vars with the square
roots of their respective weights.

For example, if you wish that y will have two times higer weight than x,
write

Compute y = sqrt(2) * y.

And run K-means.

HTH

Jan

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
OZGU GILL
Sent: Wednesday, August 02, 2006 1:31 AM
To: [hidden email]
Subject: K clustering preassigning some members in SPSS 14.0

Hi there,

  I have 110 cases and 10 variables which I want to assign into 12
clusters. I have run K clusters in SPSS 14.0 version. After checking
resuults, some of the cases would be assigned to a different cluster. It
will be more appropriate for some of the cases to be in a certain
cluster. Is  there a way to assign these cases into some preselect
clusters and then run the clustering.  Is there a way to assign weights
to variables rather than the cases. In SPSS you can give weigts to cases
there is no choice for variable that I have seen so far.  Thanks in
advance.

  Ozgu


---------------------------------
Do you Yahoo!?
 Get on board. You're invited to try the new Yahoo! Mail Beta.
Reply | Threaded
Open this post in threaded view
|

Re: K clustering preassigning some members in SPSS 14.0

Anthony Babinec
In reply to this post by OZGU GILL
I am not aware of a straightforward way to
assign cases to clusters before the fact
in K-means clustering.

In latent class modeling, there is a "Known class" modeling
capability. This feature is implemented in Release 4
of Latent Gold, available from Statistical Innovations.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
OZGU GILL
Sent: Tuesday, August 01, 2006 6:31 PM
To: [hidden email]
Subject: K clustering preassigning some members in SPSS 14.0

Hi there,

  I have 110 cases and 10 variables which I want to assign into 12 clusters.
I have run K clusters in SPSS 14.0 version. After checking resuults, some of
the cases would be assigned to a different cluster. It will be more
appropriate for some of the cases to be in a certain cluster. Is  there a
way to assign these cases into some preselect clusters and then run the
clustering.  Is there a way to assign weights to variables rather than the
cases. In SPSS you can give weigts to cases there is no choice for variable
that I have seen so far.  Thanks in advance.

  Ozgu


---------------------------------
Do you Yahoo!?
 Get on board. You're invited to try the new Yahoo! Mail Beta.
Reply | Threaded
Open this post in threaded view
|

Re: K clustering preassigning some members in SPSS 14.0

Hector Maletta
I can think offhand that the most straightforward way is conducting the
cluster analysis without the cases in question, and then assigning them at
will to the desired clusters.
Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de
Anthony Babinec
Enviado el: Wednesday, August 02, 2006 4:56 PM
Para: [hidden email]
Asunto: Re: K clustering preassigning some members in SPSS 14.0

I am not aware of a straightforward way to
assign cases to clusters before the fact
in K-means clustering.

In latent class modeling, there is a "Known class" modeling
capability. This feature is implemented in Release 4
of Latent Gold, available from Statistical Innovations.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
OZGU GILL
Sent: Tuesday, August 01, 2006 6:31 PM
To: [hidden email]
Subject: K clustering preassigning some members in SPSS 14.0

Hi there,

  I have 110 cases and 10 variables which I want to assign into 12 clusters.
I have run K clusters in SPSS 14.0 version. After checking resuults, some of
the cases would be assigned to a different cluster. It will be more
appropriate for some of the cases to be in a certain cluster. Is  there a
way to assign these cases into some preselect clusters and then run the
clustering.  Is there a way to assign weights to variables rather than the
cases. In SPSS you can give weigts to cases there is no choice for variable
that I have seen so far.  Thanks in advance.

  Ozgu


---------------------------------
Do you Yahoo!?
 Get on board. You're invited to try the new Yahoo! Mail Beta.
Reply | Threaded
Open this post in threaded view
|

Re: K clustering preassigning some members in SPSS 14.0

Steve Peck
In reply to this post by OZGU GILL
a methodological approach to dealing with such situations is provided here
 http://www.psychology.su.se/sleipner/
 (in this case, the results of a basic Ward's method cluster solution
can be used as start values for the k-means relocation analysis)

OZGU GILL wrote:

>Hi there,
>
>  I have 110 cases and 10 variables which I want to assign into 12 clusters. I have run K clusters in SPSS 14.0 version. After checking resuults, some of the cases would be assigned to a different cluster. It will be more appropriate for some of the cases to be in a certain cluster. Is  there a way to assign these cases into some preselect clusters and then run the clustering.  Is there a way to assign weights to variables rather than the cases. In SPSS you can give weigts to cases there is no choice for variable that I have seen so far.  Thanks in advance.
>
>  Ozgu
>
>
>---------------------------------
>Do you Yahoo!?
> Get on board. You're invited to try the new Yahoo! Mail Beta.
>
>
>
>
>