CHAID post-estimation

[R Allan Reese]

Sorry to appear lazy, but more to avoid reinventing the wheel (and making it square). Does anyone have any code or advice for using and displaying the e() values stored after chaid? As noted in the documentation, it produces a basic tree graph with no in-built options to modify. TIA Allan

[Andrew Musau]

Does anyone have any code or advice for using and displaying the e() values stored after chaid?

Using in what way? Displaying stored results in e() is easy. You just type:

display e(statname)

or for matrices

mat list e(matname)

where you replace statname or matname, respectively, with the actual name of the stored statistic or matrix. Note the request to identify the provenance of community-contributed commands (FAQ Advice #12).

[R Allan Reese]

Thanks but I know that! I'm wondering if someone has code for, eg, a better tree display showing numbers in each node, or a clever crosstabulation showing labels rather than values. I can do the latter for one branch but not (yet) where clusters at one level subdivide by more than one variable. It would be some sort of nested crosstabulation starting at the bottom.

Incidentally, the output includes an importance measure with the helpful comment, " Because it is possible to obtain negative decrements to fit (that is, increases in fit) due to permuting the values of a splitting variable, and due to the random nature of the permutation, the values in the "raw" row are not particularly meaningful and only the the second row, "rank" of e(importance) should be interpreted. But my dataset is showing the first level at importance 0.85 and the next variable at 0.04. I feel an order of magnitude difference must indicate more than just the 1st, 2nd.

I haven't yet looked at chaidforest but on my TODO list.