Happy New Year to you all!

I have a question on the calculation of raw propensity scores for a tree model (in this case a CHAID tree for predicting customer Churn). I've built a tree on a random sample of 60% of the population, and am analysing the distribution of rule assignments on the 40% hold-out sample. Looking at a given terminal node where the confidence level was 0.645, the calculated churn raw propensity score ($RRP-CHURN) = 0.355 for most customers who fall into this node. However, I noticed that there were a small number of customers who fulfilled the rule but whose confidence level was 0.605, resulting in a raw propensity score of $RRP-CHURN = 0.395.

I guess the fundamental question I have is: shouldn't all records who fall into the same terminal node in a decision tree have the same raw propensity scores?

Thanks in advance,

 R

 

 

Probably a question best suited for SPSS Tech Support, but at a guess I might suggest missing/null values would play a factor.

Tim