R help - Sep 2005 - "Chow Test" for classification and regression trees

Suppose one estimates a classification or regression tree (CART) for one group
or one time period; and then estimates a CART for another group or time period.
Is there a way to test for a structural change or break across the two groups or
between the two time periods, in other words, is there an analogue of a Chow
Test for CART? Has anyone ever seen anything like this or have any ideas how one
could do it? Thanks for any suggestions.
 
Charles Cameron
Professor of Politics & Public Affairs
Princeton University

	[[alternative HTML version deleted]]

On Fri, 9 Sep 2005, Charles M Cameron wrote:
> Suppose one estimates a classification or regression tree (CART) for one
> group or one time period; and then estimates a CART for another group or
> time period. Is there a way to test for a structural change or break
> across the two groups or between the two time periods, in other words,
> is there an analogue of a Chow Test for CART? Has anyone ever seen
> anything like this or have any ideas how one could do it? Thanks for any
> suggestions.
A couple of ideas could come to mind here:
  1. Just include the grouping variable (or time variable) as a potential
     explanatory variable into your tree-growing algorithm and then you
     could see whether this is picked up by the tree or not.
  2. If you've got two predictive models grown on different subsets of
     data (sorted by grouping or time) you could try to predict the values
     in the other subset for each model to check whether there are
     structural differences or not. Combining it with bootstrapping (or
     something like that) could give you an inference procedure.
  3. To do some advertising of our work: there is a working paper that
     I've written with Kurt Hornik and Torsten Hothorn about `Model-based
     recursive partitioning' that tries to combine recursive partitioning
     ideas with structural change methods that could be relevant here.
     You could fit one model tree on the whole data and then check whether
     there are instabilities with respect to the time or grouping
     variable. The paper can be obtained from
    
http://epub.wu-wien.ac.at/dyn/virlib/wp/showentry?ID=epub-wu-01_86e&back=/

hth,
Z
> Charles Cameron
> Professor of Politics & Public Affairs
> Princeton University
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
>