On Sat, Nov 10, 2012 at 1:50 PM, Gabor Oederland <[log in to unmask]> wrote:
> Thanks for the quick reply! I have some questions though:
>
>
>> The advantage of the F-test would only apply if you used a more advanced correction for multiple comparisons than the
>> standard divide by N tests.
>
> Not sure what you mean: When conducting the F-test or during the post-hoc tests?
>>> The post-hoc tests. There are several ways to do post-hoc tests (e.g Bonferoni, Tukey, etc.); however, implementing them in imaging hasn't been done before because they are step-wise procedures for each test.
>
>
>> Also, you can't limit your search region to only areas with a significant F-test because that would be considered double
>> dipping.
>
> Why not? E.g. take a neuropsychological test battery with 20 individual tests and three groups A, B, C. I would run post-hoc tests A vs. B, B vs. C, A vs. C for those neuropsychological tests only that showed a significant group effect?
>>> I'm confused. You are doing a F-test of group versus neuropsychological score OR F-tests on imaging data.
>
>
>> For these reasons, people seem not to correct for the number of multiple comparisons.
>
> But then it should be more conservative to run an F-test and limit any further post-hoc tests (uncorrected or corrected for multiple comparisons), to areas that already showed an effect? BTW, it also seems to be standard to conduct two one-sided t-tests A > B and B > A with e.g. p = .001 instead of using a corrected p = .0005.
>
>
> Leaving this aside, if I plan to run t-tests anyway, then there would be no reason to build a second-level ANOVA, or is there any? With "t-tests" inside a purely within-subject "Flexible factorial" model the effects seem to be (much) larger compared to a one sample t-test based on the corresponding con images generated on single-subject level. Or would the results of the "Flexible factorial" be correct?
Within-subject designs have only 1 group, not 3. If you have only 1
group and three or more con_ images per subject, the benefit of the
larger model, even if you don't test every relationship, is that you
have a pooled error term. This is why the larger model has different
effects. It is not universally the case that the larger model will be
more significant as it depends on the variances and covariances of
your data.
|