first of all, this is a great site and I've already learned a lot from threads here wile searching for answers to my stats questions on the web. Yet here's another one:
I understand that the independence of errors assumption in the t-test means that you shouldn't be able to predict the value of the outcome variable in any observation from the values in the other observations.
This refers to observations regardless of group membership (as in group 1 and group 2 of the t-test), right? That is, you should neither be able to predict the outcome variable based on other observations of the outcome variable WITHIN one group, nor BETWEEN the groups?
And is paired samples t-test necessary because of the specific violation of this assumption in the sense that you could predict observations in one group from those in the other?
And finally, then, the paired t-test should still require independence of observations within the 2 groups, right?
If I understand correctly, the first answer in the thread below basically confirms my presumptions, but since I'm not quite sure, I figured I'd ask about it explicitly.
Thanks and cheers, Hans