There is growing realization of the importance in randomized clinical trails (RCTs) and in risk research of understanding, not merely that a treatment or a risk factor has an effect on the outcome but specifically on whom in the population sampled does a treatment or a risk factor have such effects (via moderators), how those effects might be achieved (via mediators), and how clinically significant such effects might be (via effect sizes). Classic methods of detection of moderators and mediators have been based on statistical significance in linear models, procedures that often produce inconsistent results hard to interpret in terms of clinical significance. Methods based on non-parametric methods specifically designed to facilitate considerations of clinical significance are here introduced for binary moderators and mediators and the discussion opened for what would be needed in general.