> 2 sample t-test can take care the unequal sample size issue. but
> then do we have to be carefully choosing a proper "error term" since
> we are now looking at each 2nd order contrast independently?
No, you don't make decision on the error term, 3dttest does it for you: It is a pooled error.
> correct me if i am wrong. for a three-factor type 5 design,
> BXC(A), in 3dANOVA3, the error terms for the 2nd order
> contrasts are the followings:
>
> factorB@levelA1 has an error term of MSBC(A)
> factorB@levelA2 also has an error term of MSBC(A)
>
> factorA@levelB1 has an error term of the pooled variance of MSBC(A) and MSC(A)
> factorA@levelB2 also has an error term of the pooled variance of MSBC(A) and MSC(A)
Well, it used to be the case (and that was exactly the reason why 3dANVA2/3 were vulnerbable to sphercity violation), but with the most recent modificaitons, the error terms in 3dANOVA2/3 are basically aligned up with the same ones used in one-sample or two-sample t test.
> I am not too sure if the 2 sample t-test will be able to give me
> reasonable results because of the error issue that I am concerned
> about.
Reasonable or not is in the eye of the beholder, I guess.
As long as there are enough number of subjects in each group, and the imbalance between the two groups is not too severe, I would worry it too much.
Gang