0
$\begingroup$

I have to test the the effect of two different treatments on a set of samples. On the surface, this seems an obvious option for a two sample test. The problem is that the samples being treated are all drawn from the same pool.

Let's say I have a pooled volume of a fluid. I can draw 20 samples from this pool - 10 will be treated with method A and 10 will be treated with method B. Can I still use a two-sample test such as a t-test or a Mann-Whitney test for the two sets? Does this violate the assumption of independence of the samples because they all come from the same pooled volume, or is the fact that they are treated independently enough?

EDIT:

Adding content from comments below to help clarify the question a bit.

Sticking with the pharma theme, let's say I've collected 20 ml of blood from a single person, and I split that into 20, 1 ml samples to test two drugs - 10 samples treated with drug and 10 untreated. All 20 were derived from the same person, but tested individually. This is compared to collecting samples from 10 individuals and splitting the samples so that I have 10 treated and 10 untreated in a paired design.

I'm concerned that the original design (20 ml from a single individual) violates independence, but I'm not certain that the comparison is invalid.

$\endgroup$
2
$\begingroup$

Although they're from the same population, assuming the population is very large, the dependence is weak, if you sample them suitably of course, and the situation seems suitable for the tests you listed. Otherwise, the results of statistical tests used in many scientific papers for real experiments that have been performed about drug effect/treatment analysis would be meaningless since we always sample from the same population.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks for the response. Let me clarify a bit and stick with the pharma theme. Let's say I've collected 20 ml of blood from a person, and I split that into 20, 1MLS samples to test two drugs. All 20 were derived from the same person, but tested individually. $\endgroup$ – KirkD_CO Oct 9 '18 at 12:26
  • $\begingroup$ One more note - this is compared to 20, 1ml samples taken from 20 different individuals. I'm concerned that the original design violates independence, but I'm not certain that the comparison is invalid. $\endgroup$ – KirkD_CO Oct 9 '18 at 13:51
  • $\begingroup$ To add to @gunes's answer: as long as you want to make inferences only about that one person, then you're fine, you can use an independent samples test to contrast the two groups. If you had sampled two groups from among all of humanity, then you could do the same, assuming you wanted to make inferences only about humanity :-). $\endgroup$ – rolando2 Dec 5 '18 at 15:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.