undefined | Better HN

0 pointsnradov4y ago0 comments

The RCT is the gold standard, but observational studies can be sufficient to establish causality when the effect size is large enough with a clear dose response curve. No one has ever conducted an RCT to prove that cigarette smoking causes lung cancer, but we're pretty sure that it does.

0 comments

mlyle4y ago

> to establish causality when the effect size is large enough with a clear dose response curve.

And when there's some plausible underlying mechanism that can be studied by more rigorous means. Otherwise, observational studies can find correlation, not causation-- and they're not even very robust at that.

nradovOP4y ago

No the dose response curve alone can be sufficient. Remember that the FDA has approved multiple SSRI drugs for treating clinical depression despite a lack of understanding on the actual mechanism of action. There are several plausible hypotheses but we really don't know why they work.

mlyle4y ago

A dose response curve in an observational study is irrelevant for establishing causation.

There is a dose-response relationship between umbrellas outside and rainfall., but umbrellas do not cause rainfall.

And the SSRI point is tangential: they were approved based on RCT evidence. Observing umbrella use and concluding it causes rainfall, even though we don't know of a reason why umbrella use would cause rainfall, is quite different.

j / k navigate · click thread line to collapse

0 comments

mlyle4y ago

> to establish causality when the effect size is large enough with a clear dose response curve.

nradovOP4y ago

mlyle4y ago

A dose response curve in an observational study is irrelevant for establishing causation.

There is a dose-response relationship between umbrellas outside and rainfall., but umbrellas do not cause rainfall.

j / k navigate · click thread line to collapse