"In an RCT, if we are lucky, we find the average difference in effect produced by the treatment in the population sampled. That does not tell us what the overall outcome on this effect in question would be from introducing the treatment in some particular way in some uncontrolled situation, even if we consider introducing it only in the very population sampled. For that we need a causal model. Even less does it tell us about “side-effects” of introducing the treatment, either from the treatment itself or from our way of implementing it. These too are crucial in calculating the costs and benefits of a proposed policy. Or, as Heckman argues, suppose one wants to predict what portion of the population will experience a given degree of improvement. RCTs do not deliver that kind of result. Again, we need a causal model."