On Sun, 1 Sep 2019 00:01:37 -0700 (PDT), Cosine <
asecant@gmail.com>
wrote:
Hi:
Are there ways of conducting statistical analysis on samples taken at different times?
I figure that that is the topic for several books, sliced in an number
of different ways.
Most clinical research projects IGNORE the simple lapse of time, even
when the study takes several years. (At least, in their reporting - I
hope PIs are aware of possibilities.)
I've seen political polls reported as "first two days versus
latter two days" when the political climate was rapidly changing.
For discretely different times (as you go on to mention), I would
start by considering what is (or ought to be) considered for clinical
studies that are conducted across just a few locations or just a few clinicians. Are the subjects homogeneous? Are the results
homogeneous?
For example, collecting samples costs, and we only have some fund to collect some samples this time. Later, say, a couple of months, we have more funds, and we collect new samples. Can we combine the samples taken at different times for an overall
statistical analysis?
My experience is mainly with clinical trials that were funded from the
outside and the funders have wanted "power analyses" that show what
can be expected. Internally funded pilot studies would not be
part of the data if a protocol that was revised owing to experience
(usually the case). But I suspect that studies with too-little-power
have had funding extended, so your case is not unique. Again, there
is little harm to design if treatment and subjects (and outcomes) are
not different enough to raise suspicions about results.
Another way of thinking is that it costs for taking samples, and therefore, it seems to be a reasonable strategy to do some statistical analysis in advance to avoid continuing taking samples for a hopeless statistical analysis.
A number of major, large-scale clinical trials in medicine have taken
into account the "multiple testing" of outcomes along the way; there
are statistical procedures to follow. These can prescribe that a
study to be terminated early -- because no one should be denied
the treatment (when it is great), or because no one else should be
exposed to it (when it is bad).
In short, can we do statistical analysis on samples taken at different times? If yes, how do we do that? ...
I think I've suggested answers to those.
What are the problems of not doing statistical analysis at the end of the sample collection?
I don't know what this Q is getting at.
--
Rich Ulrich
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)