On Mon, 15 Nov 2021 22:51:50 -0800 (PST), Cosine <
asecant@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
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)