I got you. Yeah, reduced form models will have the issue of disentangling factors. This is why diff-in-diff and synthetic control models are valuable, but identification assumptions have to be clear.