Date: Fri, 24 Jul 1998 16:48:45 -0500
Reply-To: "Nichols, David" <nichols@SPSS.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Nichols, David" <nichols@SPSS.COM>
Subject: Re: Weighting in hierachical cluster?
There's no reasonable way to use a weight (that I can see) when computing
dissimilarities or similarities between two cases, which is how CLUSTER
starts. If your weights are proper integer frequency weights, you can
physically replicate the cases, so that when the analysis is done, the
duplicates will all merge together early in the solution, and the number of
them will be used in the weighted averages that result from the updating
formulas when the joining of truly different cases/clusters begins. If the
weights are noninteger sampling weights, SPSS isn't designed to handle
complex samples (and I'm not sure how that would work out in this context
anyway).
David Nichols
Principal Support Statistician and
Manager of Statistical Support
SPSS Inc.
----------
From: Jonas Gunnarsson [SMTP:fondjg@HHS.SE]
Sent: Monday, July 06, 1998 8:37 AM
To: SPSSX-L@UGA.CC.UGA.EDU
Subject: Weighting in hierachical cluster?
Dear all,
I am seeking a solution to an annoying problem. Using the hierarchical
cluster analysis module in SPSS to estimate a range of appropriate
cluslter
solutions and their respectice cluster centroids, it is as far as I've
discerned no way of weighting the data. In k-means weighting is
applicable,
however, this does not help me correct for biases in my data in the
crucial
first hierarchical analysis. It also deprives of me of about 400 extra
cases
that I could have used otherwise.
Is there a way around this problem, such as constructing new variables in
the raw data file using the sample weights and THEN applying the
clustering
procedure to calculate distances? Any suggestions would be much
appreciated!
--------------------------------------
Jonas Gunnarsson
Foundation for Distribution Research at the Stockholm School of Economics
fondjg@hhs.se
Internet: www.hhs.se/fdr/staff/jonasg.htm
| 
