topkatz@MSN.COM sagely replied: >I guess you could use clustering for this, but it looks to me more like a >straightforward application of Link Analysis / Associations Analysis / >Market Basket Analysis.

Me too. Of course, you can pick whichever name you like. They're pretty much the same thing, with different wrappers. Link Analysis typically means you're doing the same basket analysis, but drawing a really cool picture to go with it. :-)

>Here is a nice presentation on Market Basket Analysis that introduces the >basic concepts of support (percentage of all combinations in which a >specified combination occurs), confidence (probability that if part of the >combination occurs, then the whole combination occurs), and lift (ratio of >actual confidence to confidence if subevents were independents) and >discusses some more advanced analytics that can be added to it. > >http://nymetro.chapter.informs.org/prac_cor_pubs/Ausleder-On-market-basket-analysis-May-04.pdf > >I'm not sure what kind of code is available out there in the public >domain, but it's quite doable with some SQL and array manipulation.

I would suggest PROC MEANS with an emphasis on the COMPLETETYPES and CHARTYPE options. You may have to prep your data first, so you can run it through the proc, and you would need to write your own 'analysis' code to compute stuff like the lift for your link analysis. (Not you personally, TMK, just a generic 'you' out there.)

>(SAS Enterprise Miner has an Associations node, but that may not help you >much.)

Well, if you've got it, flaunt it. :-) The Associations node does do 'market basket analysis'.

David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

_________________________________________________________________ On the road to retirement? Check out MSN Life Events for advice on how to get there! http://lifeevents.msn.com/category.aspx?cid=Retirement