Comment by n4r9

Even when you lack this level of control it may still be possible to infer causal effects. There's some interesting work that I've seen on this, good write up here: http://www.michaelnielsen.org/ddi/if-correlation-doesnt-impl...