It is remarkable how often when you penetrate deeply into a non-trivial, technical domain you discover how tenuous its practitioners’ understanding is of the essential causal dynamics of the domain. Similarly, you often discover that they have, sometimes unknowingly, largely given up on causal claims, and are mostly engaged in post-modern statistical games.
I mentioned this recently in the context of drug discovery and mechanisms of actions, but it applies much more broadly—and it matters. When you don’t know whether x → y, or y → x, or maybe sometimes x → z → y, you’re only one bad data set from thinking maybe wet streets cause rain after all. This applies to some technical practitioners as well, who regularly demonstrate they have successfully prevented people from realizing they don’t know the core causal elements of their domain, sometimes because there aren’t any. I’m thinking of golf pros, financial advisors, and, yes, epidemiologists. All of them regularly prove that wet streets cause rain.