I want to study if atheists/non-believers are having a harder time gaining results from 12-step facilitation treatment for alcohol dependence. My hypothesis is that a higher degree of belief in god/higher power is correlated with a higher degree of sobriety among patients going through a 12-step treatment for alcohol dependence.
I already have all the data I need. A group of around 200 ppl answered a big questionnaire at the beginning of treatment, giving a measure of self evaluated degree of "spiritual acceptance" (combined result from several items). I also have a binary result (sober or not sober) 1.5 years after treatment. So I have an independent variable which is a floating scale and a dependent variable which is binary.
My question is what would be the best strategy for analysis of the data. Since my question is whether atheists are disadvantaged, I'm considering to split the patients in two groups: one group with those who are beyond one standard deviation on the lower end of the spirituality scale and one group with the rest. In other words I would create an arbitrary division and have one "atheist" group and one group with those who are at least somewhat spiritual, then maybe do a chi square test. I'm not sure if this is considered good practice though? Would it be better to do a logistic regression and leave the dependent variable as a floating scale?