For reasons I won't go into, I want to test for group differences between an entire sample A ($n=65$) and a subsample B ($n=45$). All of the variables I want use for mean comparisons are continuous.

Is there an appropriate test for this?

This doesn't seem to meet the sample independence condition for the t-test or Mann-Whitney U. Is it better to just test for group differences using a t-test between B (n=45) and !B (n=20)?

  • $\begingroup$ Logically if the mean (or other location measure) of $A$ differs from $A+B$ then it differs from $B$, and vice versa. The same applies to a more general U test, even one comparing $P(A<B)$ to 0.5. Rejection of one form of hypothesis implies rejection of the other. If you did form something like a t-test for comparing dependent samples like that, you'd end up with something that was in effect doing the $A$ vs $B$ comparison anyway. $\endgroup$
    – Glen_b
    Sep 27, 2013 at 0:12

1 Answer 1


Your last thought is the right one. Test between two sub-samples, not between a sample and a sub-sample. In this, a t-test might be appropriate, if it's other assumptions are met.

The only way I can think of to test between a sample and a subsample would be simulation, but that would wind up equivalent to testing between two sub-samples.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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