Hacker News new | ask | show | jobs
by chairhairair 106 days ago
19 participants.
4 comments

>> therefore 19 participants completed the study (2 females and 17 males) and their data are presented throughout

Who in their own mind decided that this is a "study" worth publishing?

You're reading the study wrong.

You read

  We saw this effect, so it's real. 
In actuality it is

  We saw this effect in a small study, so it's worth doing a larger study.
It's worth publishing because it's evidence and motivation to do further studying. And if you're asking "Why not start large?" the answer is obvious: money.
Especially in dietary studies. You either spend a lot on high quality, controlled studies where you can nail down parameters (takes a LOT of labour), or you spend on facilitating much larger studies where you make up for precision and control with volume.

There are trade offs in either case and some types of research where one is more suitable than the other. But the best case is a combination of the two, and it's exceedingly rare.

Maybe there are other options but this seems to be the polar nature of these studies from what I've seen.

> We saw this effect in a small study, so it's worth doing a larger study.

That never happens.

That doesn't change the meaning.

Reading it wrong tends to make that happen too. Since people read it as fact they end up arguing "why fund a more expensive study to show the same thing?"

A lot of this stems from not understanding how science actually works. How there's never proof, only confidence bounds

The paper includes a section on power analysis which justifies the sample size (although the justification is for a sample of 20, they recruited 25 eligible participants and lost 6 in screening).

Some points though:

- A within-participants study has inherently more power than a between-subjects study. Trying two different diets with the same person removes a lot of variables that you'd need to control for in between-subjects studies (and yes, they randomized the order of intervention and found no difference based on order)

- It looks like this was conducted in a way that supported compliance with the protocol, and using analysis techniques that would be unwieldy for a much larger sample size.

Even with N=19, the reported significance is very compelling.

Someone with quotas
Average age was 57, which may be rather high. Also: why not test out combining both diets?
That was the point I stopped reading.
The number needed for a study to get significant results depends on the strength of the effect it is measuring.

For example if I have a bag full of thousands of coins, pull out 19 at random and flip them sequentially, and they all come out heads I'm going to conclude I have a bag that is overwhelmingly coins that are heavily biased toward coming up heads.

Are you going to say my sample size was too small to support that conclusion?

To see if their sample size was too small you need to at least read the part where they do the math.

True, or you can have a prior about how many people you need to get a significant result in a certain area.

I have one for "diet changes have large impact" and the number is different enough feom 19 that I'm happy to stop reading at that point.

go on, explain why you think there is a problem with the sample size. But no words, only clear statistical calculus. I'll wait.