I will try to be as detailed as possible. I'm looking to build a regression model for the independent variable 'Success' against a variety of characteristics, one of which is Age.
The relationship between Age and Success is non-linear, the R2 for a trend line of the scatter graph for X2 is higher than that of the R2 for linear (and logisitic is somewhere in the middle). I'm aware that you can't use logistic regression when the relationship is non-linear and you need to transform into a linear variable.
I have transformed using; Age^2, log(age), 1/Age (each time not transforming the dependent variable) and run the regressions again. The 1/Age R^2 statistic for linear regression is over 2 * that of the unmodified Age characteristic, however it is an improvement from 0.0017483 to 0.003289.
I've then looked at the Weight of Evidence table (LN[(Success(t)/sum(success))/(not success(t)/sum(not success))] for each banding) and I still can't get a linear relationship.
Am I missing a trick by not transforming the dependent variable? If so, 1/dependent won't work as they are 1 (success) and 0 (not success) or is there another transformation I could try which could get a better linear relationship?
Any/all help or comments would be appreciated.