I would like to use Mplus to test indirect effects of childhood abuse/neglect through a continuous mediator to a dichotomous outcome. Neither the exogenous predictor nor the mediator is randomized. Reading Muthen (2011) and Valeri & VanderWeele (2011), it seems that this analysis may not meet the assumption for causally defined effects. Am I interpreting this correctly? I would prefer to report indirect effects to the observed rather than the latent variable, but not if it violates key model assumptions!