08-29-2013 10:01 AM
I would like to use some continuous variables to see if I can predict a certain behavior (such as "fallers-people who had a history of falls" versus "non-fallers").
My dependent variable is "Group" (0=non-fallers, 1=fallers).
For the independent variables, I use couple of continuous measures.
Since the absolute values of my independent variables would affect the "unit odds ratio" (although the "range" odds ratio would be the same), how should I choose the "unit" of the independent variables (such as "meters versus centi-meters")?
For example, If I input my data in centi-meters, I would have the unit odds ratio as 0.6. But if I input my data in meters, then I would have the unit odds ratio as 0.002.
Also, if I would like to compare the unit odds ratio among different independent variables, how should I do to make it fair since their absolute values are very different due to different scales/units?
Moreover, if the distribution of a certain independent variables is not normal, should I transform my data first (such as log transform) before running the logistic regression?
thank you in advance!
08-30-2013 05:16 AM
why not standardizing all scales to mean 0, variance 1 or even keeping the mean and shrinking / expanding all to the same variance? Then you should get the same beta for centimeters and meters. And sure, for skewed distributions apply the log transform before.