News for hackers. Stuff that matters.
Re:Experimental depression treatment is nearly 80% effective... by herodoturtle on Wednesday December 29, 2021 @09:12PM (#29732716)
Thank you for saying this.

More n is required, yes, but small n results like this are a noteworthy step in the right direction.

There is often good reason why n is small. Statistically significant results like this one pave the way to future experiments with larger n.

Re:Experimental depression treatment is nearly 80% effective... by jcims on Wednesday December 29, 2021 @10:10PM (#29733090)
I had an occasion to deep dive into hundreds of clinical trials about a year ago, talk to PIs and trial participants, etc. One thing I didn't even have awareness of, let alone appreciation for, was how essential it is to have a good trial design (or maybe how an insufficiently considered trial design can be vulnerable to catastrophic failure modes).

It seems that running these 'small batch' studies/trials can be tremendously helpful to inform and refine the approach before executing on a larger scale. Of course if it takes 3 years to reach an endpoint it probably doesn't make sense to serialize them, but to my layman perspective there appears to be clear value (and in this case encouraging results).

Re:Experimental depression treatment is nearly 80% effective... by esyir on Wednesday December 29, 2021 @09:31PM (#29732853)
I'd also like to chime in on "effect size matters". After all, you don't need particularly large N to conclude that decapitation is going to be fatal.
Re:Experimental depression treatment is nearly 80% effective... by mlyle on Wednesday December 29, 2021 @09:34PM (#29732871)
Yes---

Small n studies are often the most exciting. If you get a p<0.01 effect in 10 people vs. 10000 people... the lower bound of the effect size is similar, and so for the same prior each study should change our belief by a similar amount... but the upper bound of effect size is much larger in the large study.

The only thing that messes this logic of mine up a bit is publication bias. There's a much larger chance that small studies sit on the shelf unpublished if there's no clear effect.