The problem is that Y and Z are reciprocal, in that Z is a predictor of Y and Y is also a predictor of Z. In my paper, I am attempting to model the effect of X on Y and then effect of Y on Z. This is a non-recursive model that I believe Mplus can handle.