proc factor with polychoric correlation matrix: yields errors

[mszommer]

Sure, will run it with fewer Obs. and with the simpler code you recommended.

The idea was to reduce the no. of variables (a mix of only ordinal and nominal) and then run a cluster analysis based on the factor scores. However, is it recommendable to name/profile the factors that resulted from a principal component analysis? OR would you have recommended running a cluster analysis directly with the 103 variables?

[Ksharp · Accepted solution]

No. You have ordinal and nominal variables, not continuous variables.Therefore you can not use PROC CLUSTER or a principal component analysis directly. Best choice is PROC PRINQUAL .

[Ksharp]

Or you could try PROC VARCLUS, but I don't know if it could fit your ambition .

[mszommer]

Hello KSharp,

I did use proc factor (method=principal) using the output from proc prinqual, in order to reduce my variables.

I then wish to segment the respondents; would you still recommend proc varclus or k-means?

Regards

MS

[Ksharp]

You mean segment the obs ? I don't think proc varclus or k-means can do that. both need variables be continuous.

[mszommer]

Yes, I do. The very aim was to segment the survey respondents based on the IVs. Since the nature of the IVs were varied (both metric and non metric), I was proposed to used proc prinqual.

Is there any other means to segment the customers? I use one of the questions as the DV (how often they buy from us: nominal)

[Ksharp]

Can you get those two engvector after principal component analysis?

Plot these obs by these two engvector. And may be you could find which obs be closer.

[Ksharp]

Here is an example. 

http://blogs.sas.com/content/iml/2014/11/07/distribution-of-blood-types.html

Check the last picture.