Hacker News Reader

Comment by pjmorris | Hacker News Reader

I am having trouble accepting your argument. Judea Pearl presents [0] the debate over whether smoking causes cancer as an illustration for why correlation is not sufficient. Two plausible, contradictory, models ('smoking causes cancer', 'no it doesn't, something else does') could not be settled by observing correlations. The debate was resolved through the addition of an explanatory mechanism, and a test of the mechanism's influence. Pearl, and others, have worked on the mathematical machinery necessary to examine this kind of causality.

[0] http://singapore.cs.ucla.edu/LECTURE/lecture_sec3.htm

View on HN

none of this applies to a double-blind, placebo-controlled, randomized clinical trial (which is what this article is).

the reason it's the gold standard for medicine is because the only thing that is different in aggregate between the groups is the intervention. Therefore, it is the strongest possible evidence of causality we have, and does say something about causality (rather than correlation).

It's important to replicate, but the points made by Pearl are not relevant here and the mechanism is moot when you have a causal link shown by RCT.

Andrew Gelman has a nice blog post on the shortcomings of the double blind clinical trial:

http://andrewgelman.com/2018/01/08/benefits-limitations-rand...

To be honest, I think a randomised clinical trial is more of a starting point than an endpoint in terms of 'knowing' that an effect is genuine. The 'gold standard' moniker overdoes the authority of the RCT a

Unfortunately, we don’t understand the exact mechanism for many (most?) available medication.

Correlation != causation. Which is why you need to do randomized testing and try and account for other causative factors. But if you give a bunch of people a medication and they stop dying, you can pretty much say this is valuable and start giving it to people.

The issue that you're raising does not apply for randomized treatment.

This study established causality with double blind test and thee effects size was really big.

Reply on news.ycombinator.com