I’m personally a fan of holdout groups because of their accuracy, although they’re pretty difficult to implement. I know you’re a causal inference guy — do you have recommendations?