The FastDay Forum

Hi Coffecat,

"N=3 or even 5 is definitely not acceptable in "most experiments"

Whilst I fully agree that 25, 50 or 100 or more would be way better, there are many publications that feature stats analysis on the results with low 'n' values. For interest, I looked at 'n=3' and 'Fasting' on PubMed and get more than 400.

BW.

I went and did the same thing, but did you look at the abstracts in question? (n = 3,120) n3 fatty acids, n-3 PUFA, 171 participants represented 6 major haplogroups: L0 (n=78), L1 (n=3), L2 (n=30), L3 (n=53), L4 (n=1) and L5 (n=6), etc.
I think you will also find most studies with n=3 are either not in high quality journals, not recent, and probably not dealing with humans but with cells or genes. Any human study with only 4 study is at best a pilot study for something really rare and exotic.

Its a statistical given, studies with a variable with large variation and loads of confounding factors (unlike genes, unlike controlled laboratory situations) need a larger sample in order not to end up with a fluke positive finding. If it wasn't like that, my life would be easier...

Hi Coffecat,

"need a larger sample in order not to end up with a fluke positive finding"

So in our situation, as pertaining to the progress tracker data, what would you consider sufficient sample size. That is, what's a minimum acceptable value for n?

That depends on the design (who and what do you want to compare), the standard deviation of the variable that you want to observe (should be available from the progress tracker but not to me), and size the of difference is that you want to be able to detect (i.e. 10 grams or 1 kg/week weight loss).
In addition, what other variable do you want to control for (gender, age, BMI, adherence, physical activity?). In linear regression you need at least 10 obervations (= participants) per variable. Don't know about logistic regression (your design determines which statistical method to use). You can avoid adjusting for these by looking at obese sedentary middle-aged women only, but then you have to recruit/select these first.

I'm not trying to put you off, but this is how you usually calculate your sample size. Doing a study on a sample without enough statistical power is considered spurious.
If you give me a design and the standard variation of the outcome/effect that you want to study, I might be able to give you an indication.

The value of n usually kills you in the significance tests, the statement made in the Subject line of this thread is very unlikely to be true at say 95% probability.

As a starter for 10 you can calculate the mean and sd of two sets of data and see how much overlap there is in the +/- 3sd range. If the overlap is huge, you need a bigger population, less variability or a bigger difference in the average (longer trial ?) to help bring it into significance.

[carorees]

I wonder if Moogie could add the SD and range data to the reported mean and median in the progress tracker...that would tell us quite a bit, no?