Suppose we have outcome variable Y and two predictors X1 and X2, where X1 and X2 sort of come from a type of category. For example, X1 is the relationship score with friends and X2 relationship score with family. Not the same, but they reflect some thing together (social relationship, in this case).
I want to set up a model to test the hypothesis that the diversity/balance of predictors can influence the outcome. Some thing like this: Suppose there are two persons who both have scores of 10 on X1 and X2 total. I suspect that the one with score of 5 on X1 and 5 on X2 will have better outcome than the one with 10 on X1 and 0 on X2.
I think a normal model (Y = X1 + X2) or an interaction term (Y = X1 + X2 + X1*X2) seem not to do the job. Maybe there needs to create new variable(s) but I'm not sure which to create.