Doaa,

I suggest you look into the "counterfactual" approach to causal inference. A Complier Average Causal Effect (CACE) model is fairly easy to implement in this case, and will improve inference over the classic Intent-To-Treat analyses.

Specifically, I suggest looking into the effect of response, not assignment, on your outcomes, using assignment as an instrumental variable. In a randomized trial, this usually works well.

Pat

Thank you too much

How can I apply on SPSS?

Doaa,

I don't use SPSS, but an instrumental variable can probably be applied in AMOS. The most straightforward SEM application in this case would be regress Response on Assignment; Outcome on Response, but NOT on Assignment; and covary the disturbances between Response and Outcome. The path between Response and Outcome is then the parameter of interest.

Pat

Thank you too much

Can I use binary logistic regression (non responders versus responders)?

Doaa,

Yes, as long as your SEM package handles binary endogenous variables, which most do these days, the logic will still apply.

Pat