What happens if you treat the DV as continuous?
   
  There is also work by Marilyn Ritchie at vanderbilt (I assume genetics) and Rob Culverhouse (cced on this) in my group that uses a partitioning algorithm (not tree but perhaps usable) you could try.
   
  Bill Shannon

Peter Flom <[log in to unmask]> wrote:
      I would be interested in any references anyone can supply on ordinal trees, either alone or in combination with ordinal logistic regression.

Here is a brief outline of what I am trying to do:
We have a DV that is ordinal - level of dementia in the elderly.  In our data set, it has six levels, with more people in the middle levels than the extreme ones.  Total N is about 1,000.  We have a great many potential IVs (almost 2000) but many of these are highly correlated, and some are more likely to be related to the DV than others.  I've done a lot of data reduction, getting it down to about 100 IVs. 

The problem is that the relationship between the DV and the IVs is different at different levels of the DV.  For instance, some IVs are similar at DV = 1, 2, or 3 but then jump and are similar at 4,5, or 6.  Others show different patterns.

I've tried a few different things.  One that seems to show promise is first doing a tree of 1,2,3 vs. 4,5,6 then doing trees among 1,2,3 and 4,5,6 separately.  But this is problematic because the first tree, while it works fairly well, does not work nearly perfectly. 

I am using CART for the tree analysis, and have SAS and R for other statistical analyses.

So, before I reinvent the wheel, I wanted to ask if anyone has seen something like this before.

Thanks in advance

Peter

Peter L. Flom, PhD
Brainscope, Inc.
212 263 7863 (MTW)
212 845 4485 (Th)
917 488 7176 (F)



---------------------------------------------- CLASS-L list. Instructions: http://www.classification-society.org/csna/lists.html#class-l

----------------------------------------------
CLASS-L list.
Instructions: http://www.classification-society.org/csna/lists.html#class-l