Dear allstaters
A logistic regression with several orthogonal independents yielded no
significant coefficients, but a tree analysis using Rpart in Splus gave me
some nice results with one of the variables being significantly associated
with my outcome after the tree introduced a breakpoint at the 62nd%ile. I'm
assuming the null result with the logisitic regression occurred due to a
lack of linear correlation on the logit scale between my outcome and any of
the independent variables, which were standardised beforehand. Or am I
overlooking something?
cheers
Mark Booth
|