Thank you for answering.
Now I understand the reason behind that. But how can I investigate
interactions amongst conditions in 2nd level without loosing the
statistical power this way?
The one-sample (and paired t-test) show activity without any doubt
(FWE corrected, k=5). If I use ANOVA, instead, it will simply
disappear even at p < 0.001. Previously I thought setting
"independence = no" would be the same as doing the contrasts in 1st
level. Apparently I was wrong. Can we conclude that ANOVA is producing
false negatives and stick to analysis without interactions? What would
you do in my case?
Thank you.
Dorian
2009/10/16 Lily Liu <[log in to unmask]>:
> When you contrast A > B in 1st level and make one-sample t-test in 2nd level
> , you treat A and B as within-subject conditions. But if you contrast A and
> B separately in 1st level and bring them to two-sampled t-test, you treat A
> and B as the between-subject conditions, which bring more variations and is
> less sensitive.
>
> Li Liu
>
>
> 2009/10/16 Dorian P. <[log in to unmask]>
>>
>> Dear all,
>>
>> I am getting confused with results from different tests. If my
>> understanding is correct, there should be no difference if I:
>> (1) contrast A > B in 1st level and make one-sample t-test in 2nd level OR
>> (2) contrast A and B separately in 1st level (contrast "1") and bring
>> them to full-factorial (ANOVA) or two-sampled t-test.
>>
>> Well, this is not the case. Two-sample and full-factorial are
>> producing different results (even when I set dependent measurements). With
>> one-sample t-tests I find strong activity (FWE corrected), which
>> disappears if I follow way 2. Only the paired t-test produces the same
>> identical
>> result as the one-sample t-test.
>>
>> Can someone explain the differences between these tests?
>>
>> Does this suggest it's better to do many one-sample t-tests instead
>> of a single ANOVA in 2nd level?
>>
>> Do ANOVA really loose statistical power (maybe because of degrees of
>> freedom)?
>>
>>
>> Any help is welcome. Thanks in advance.
>
>
|