Hacker News Reader

Comment by sz4kerto | Hacker News Reader

Good idea, but a hard limit like 10000 is not that good. Statistical methods don't require specific sample sizes. Cherry picking, incorrectly chosen methods are the problems, not the sample sizes, usually. A hard limit would suggest that anything above this sample size is credible, while that's very far from the truth. The reproducibility crisis is very much unrelated to sample sizes.

View on HN

Sure, you can't design a study badly and make it up on volume. — but there's a point at which the sample is so small we really don't need to look much closer.

For instance, I don't want to spend time figuring out whether an n<100 study is any good. It isn't.

Surely the required sample size depends on the effect size. You won't need n>100 to learn that elephants are larger than ants. The problem is in experiment design and interpretation generally, not sample size specifically.

That said, all of us who know how to read a scientific paper (and have institutional access to the journals) should take 5 mins and point out flaws when we see them.

You don't need a large sample to establish that beheading is fatal. Sample size is only one factor in statistical significance; often it's a fairly unimportant factor.

> For instance, I don't want to spend time figuring out whether an n<100 study is any good. It isn't.

So an AIDs cure that cured 80 out of 90 men wouldn't be worth reporting on HN?

Studies that have >10,000 people I'd think are close to non existent, these are hugely expensive and I think most are well known.

I think it's a job for the scientific community to work out, I can't see moderators on HN doing it.

Personally I wouldn't click on a link that was reporting on a potential AIDS cure that was tested on 80 or 90 people.

I'd click on a link on a report that cured 80 of 90 AIDS patients.

I think you've misunderstood the nature of preliminary clinical trials.

Reply on news.ycombinator.com