Gang Wrote:
-------------------------------------------------------
> > What accounts for the difference?
>
> You're seeing the differences because of the
> respective null hypotheses those two tests are
> constructed against. The null hypothesis for the
> first test is
>
> H_0: stim1 + stim2 = 0
>
> and the second is
>
> H_0: stim1 = 0 and stim2 = 0
>
> > Essentially, we want to obtain a Full F,
> parameter estimate and corresponding
> > t-value for the contribution of both stim1 and
> stim2. Scenario 1 appears to be
> > what we want, no?
>
> The phrase of "the contribution of both stim1 and
> stim2" is very vague. The first test of yours
> looks for the combined (or average) effect of the
> two stimuli (its t- and F-statistic are
> essentially the same thing), while the second test
> (F-statistic) gives you the effect from either of
> the two (i.e., when the null hypothesis gets
> rejected), but does not differentiate between
> them. In addition, 3dDeconvolve also provides the
> individual effect significance (t-statistic) for
> each of the two stimuli in the second test, which
> are just duplicates for the two regression
> coefficients in this case. To me the second test
> is usually much more informative and interpretable
> than the first one, but there are exceptions
> (e.g., when testing for main effect).
Gang,
I wanted to revisit this. when you the mention the exceptions where the 1st hypothesis is informative, e.g. when testing for main effect, could you clarify what you mean by that?
Thanks,
Michael