Q the least number of samples

[Cosine]

Hi:

What is the least number of samples that could still support practically meaningful statistical analysis?

For example, would 3-5 samples be enough? Or at least 10 or else?

[Rich Ulrich]

On Mon, 15 Nov 2021 22:51:50 -0800 (PST), Cosine <ase...@gmail.com>
wrote:

>Hi:
>
> What is the least number of samples that could still support practically meaningful statistical analysis?
>
> For example, would 3-5 samples be enough? Or at least 10 or else?

Is it a "meaningful statistical analysis" to declare that you have
observed something unique, never seen before?

Most data points comes as measurements that are rather
familiar. I remember a lab researcher who was applying
principles of /probability/ (if not statistics) when he did
testing on three samples -- One was Control in order
to confirm that everything (procedures, chemicals, etc.)
was working as expected in the null case, and two were Test
so that the hoped-for result (3 SD difference from expected)
was replicated and not a fluke of some procedureal screw-up.

If "statistical" requires making use of internal variation among
the samples on hand, you can look at p-levels that are possible
to achieve for separate tests. A t-test with 1 d.f. requires a
huge numerical difference, and (perhaps) a "meaningful" result
requires having great faith that a number of assumptions have
been met.


--
Rich Ulrich