We will very appreciate if you can further advise us the following questions:
2012/6/15 MCLAREN, Donald <[log in to unmask]>:
> On Fri, Jun 15, 2012 at 4:57 AM, <[log in to unmask]> wrote:
>> We are performing a VBM study. We have 2 groups (say patient and control)
>> and 2 phases (pre- and post-training) to test. So we build a 2x2 ANOVA using
>> flexible factorial design, with 3 factors (Subjects, groups, and phases) and
>> 4 possible covariates.
>
>
> Do not include covariates in the model as you have repeated-measures.
Can you explain a little more about this? If we can't include
covariates in our model, then how can we control the global variates
like total gray matter volume and age?
>> We search the mail-list archive and found that only the group*phase effect
>> and phase effect are valid. The group effect should be tested using 2-sample
>> t-test after average 2 phase data. Thus, is it valid that we compare the
>> phase difference within single group? Or it is better to set another paired
>> t-test design for this kind of comparison?
>
> When you test T1 vs. T2 in a model with 2 groups, then the contrast for T1
> vs. T2 for a single group uses the pooled estimate of the variance across
> all subjects. This is fine, but will be different than a model with only a
> single group.
Is it valid to report the results from different models (say T1 vs T2
from paired t-test and difference of difference from 2x2 ANOVA)? Or we
better to report the results from a single model?
--
Best regards
--
§ù¬F©þ
Cheng-Hao Tu, PhD
Post Doctoral Research Fellow
Integrated Brain Research Unit
Department of Medical Research and Education
Taipei Veterans General Hospital
No.201, Sect.2, Shih-Pai Rd.
Taipei
Taiwan
Email: [log in to unmask] or [log in to unmask]
Tel: +886-2-28712121 ext 3474
Fax: +886-2-28745182
|