Dear SPSSers
Have been using classify > tree to predict a numeric response variable from 21 similar numeric response variables and 2 categorical variables.
Need help interpreting IMPORTANCE using CRT method for growing trees
the response and its 21 similar variables are probabilities.
On the raw probabilities all raw importance values are low< .1
BUT using lgt(prob) raw importances are much higher, highest being about .3
However the regularised variances as percentages are very similar, so here are queries
1. what regularised % importance are likely to be useful? am considering 75% as cut off
2. are there criteria for raw importance?
3. why does transforming variables monotonically make such a large difference to raw importance?
References to any material available on net would be gratefully received.
ps. why is split file forbidden for classify, tree?
best
Diana
_______________
Professor Diana Kornbrot
University of Hertfordshire
College Lane, Hatfield, Hertfordshire AL10 9AB, UK
+44 (0) 208 444 2081
+44 (0) 7403 18 16 12
+44 (0) 170 728 4626
skype: kornbrotme_______________________________
Locked
