Resampling for Binary Logistic Regression

hhhh

New Member
#1
I have 9 independent predictor variables that affect 10 independently tested binary outcome variables in SPSS 16.0.

There are about 2000 cases, and while many of the outcome variables are nearly evenly distributed between outcomes 0 and 1, some of the outcomes are considered "rare", where 1 occurs only 10% of the time and 0 occurs 90% of the time.

This ends up in a classification table that can correctly predict a 0 outcome up to 95% of the time, but correctly predicts a 1 outcome only 20% of the time. It is important to the research that each model reliably predicts both 1 and 0 outcomes

It seems necessary to somehow re-sample the cases where outcome=0 so that outcomes of 1 and 0 are nearly equal.
Can this be done efficiently in SPSS? Is there a command or protocol somewhere for this?


Thanks in advance for any help with this!!!