Hacker News Reader

Comment by cjohnson318 | Hacker News Reader

"Four men and four women participated in two trials"

I'm not a medical researcher; is this a normal size for a study?

Say you are trying to distinguish between a placebo and a cyanide pill; test subjects are young rats (base rate of death extremely low). How large of a sample size do you need? I think one control, one test subject is enough to be certain, because the base rate of death is insanely low, when measured on a per-minute basis.

Sample sizes can be small and still very significant when the effect is large.

I would want about four of each rat still. The control could die of something else...

I'm quite confident the only reason to use four rats in a study of cyanide is that you really want to kill two rats.

Yeah, because you already have a very well-established prior on the effects of cyanide.

> when the effect is large

You mean when the effect is guaranteed. Cyanide is reasonably guaranteed to have the expected effect every time which is why you don't even need the control.

With only 4 test subjects you can't really draw a strong conclusion.

Maybe it's a relatively quick and cheap way to decide if it's worth doing larger and longer studies.

Voice of reason

Medical research is expensive to conduct, so it's not uncommon to see N<20. Additionally, more data is not necessarily better data since humans lie and it's unethical to force their compliance, so there's more of a risk for bad data with large trials. And throwing out bad data is selection bias.

It's enough for such a large effect, but obviously it doesn't tell us anything about the long-term impact of trying it as a lifestyle.

There are two problems: "No differences (p>0.05) between interventions were found for plasma insulin or glucose AUC." That can very, very much be explained by lack of power. But the main problem is that a group of 8 people has a very low probability of generalizing across the population. And of course, very few people sit 8 hours without moving.

That said, it's more than promising enough for further research.

I thought the exact same. I honestly think this post should be deleted on that fact.

Reply on news.ycombinator.com