Re: st: clustering with a new dataset

[Frank Gallo <fjgallo@mac.com>]

<>
<>

Hi Walt,

I am a Stata beginner so I have little to offer you regarding  
procedures available in Stata: maybe other listers can. However,  
conceptually speaking, it sounds as though she wants to cross-validate  
her model. A well-fitting cluster (or factor) model is tentative. It  
requires post-hoc model validation. The researcher may use a random  
sample from a validation holdout sample for cross-validation (i.e.,  
within sample replication, which requires a large sample size). Though  
even if a model fits the data well, it does not mean that it is the  
correct model or even the best model to explain the phenomenon of  
interest. There may be equivalent models that fit the sample data or  
other data sources equally well. If the researcher uncovers equivalent  
models, there is no statistical technique for discriminating among  
them. Only on substantive knowledge about the phenomenon can the  
researcher decide which equivalent model is best. The researcher may  
judge a model "good" on both theoretical and statistical grounds, and  
thus, provisionally accept the model. Cross-validation procedures on  
different independent samples (seems like your case) from the same  
population can enhance the utility of the model. You may compare the  
models by examining the overall fit indices (e.g., chi-square, RMSEA)  
and the significance of path coefficients to offer the client some  
insight. I hope this helps.

Best,
Frank



On May 3, 2009, at 9:08 PM, Data Analytics Corp. wrote:

Hi,

I ran a cluster analysis last year for a client using "cluster ward  
varlist" where the variables in varlist came from a survey.  This  
worked fine and the client was happy.  This year, she returned with a  
new dataset (same variables, just new values from a new survey) and  
wants last year's clusters applied to this year's data.  I can't see  
how to do this - in fact it doesn't seem to make sense.  Any  
suggestions, or should I tell her that I can just rerun the old  
commands and MAYBE the same clusters will appear?

Thanks,

Walt



--
________________________

Walter R. Paczkowski, Ph.D.
Data Analytics Corp.
44 Hamilton Lane
Plainsboro, NJ 08536
________________________
(V) 609-936-8999
(F) 609-936-3733
dataanalytics@earthlink.net
www.dataanalyticscorp.com

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/