[danharaj]

That doesn't tell you how to judge effects in specific situations. You can infer lots about a population without being able to particularize it at the level required to implement policy. Twin studies are observational. They aren't experimental. Twin studies can never give you a model of how genetics influence intelligence, unless you want to do extremely unethical experiments on twins.

It is akin to the distinction between a non-constructive proof and a constructive proof of a mathematical property of a class of objects. The former lets you work under the assumption that that property holds, but if you want to work algorithmically you need a constructive, computable proof.

Likewise, statistical inference can give you interesting signals to pursue, but it doesn't give you the models required to understand how those signals come about or how to influence them. The idea that they do is at best folly and at worst malice.