I would be grateful for some advice. I have a binary response variable and
several categorical "predictor" variables. In initial data screening I used
two-way contingency tables between the response and each individual
predictor and observed several significant effects. I then ran a logistic
regression involving all the variables and no predictor had a significant
effect. Is this a manifestation of Simpson's paradox, and how should I
proceed with the analysis - use the logistic regression and say the
predictors have no effect, or use the selected results from the two-way
tables (which I must admit does not feel very satisfying to me). Are there
any references documenting this (presumably common) problem?
Thank you,
Maged
_________________________________________________________________
Send and receive Hotmail on your mobile device: http://mobile.msn.com
|