(2) Group effects in the presence of an interaction. In the presence of an interaction, I create a mask that excludes voxels with a significant interaction. Then I either use the same model and look at the group effects in voxels that didn't have an interaction or use the simpler model to gain 1 degree of freedom. Both approaches are statistically valid. The key is to exclude the covariate interaction voxels from the analysis of the group difference effects.
(3) In writing up the paper, you should state the group differences are based on the covariate-adjusted means. This terminology isn't routinely used, but I feel strongly that the field should start using this terminology to increase the transparency of the statistics reported in papers.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.
On Fri, Jan 13, 2012 at 10:18 AM, Torsten Ruest
<[log in to unmask]> wrote:
Excellent, thanks very much for this Donald! I ran some quick OLS models (will do the FLAME over the weekend):
1st model, includes only the 2 groups, ie patient (EV1 is the cope obtained from after-before) vs control (EV2: after-before)
contrast: patient [1 0] shows some differences
contrast control [0 1] is empty
contrast: patients - control [1 -1] shows some differences indicating that our patient group has higher activation after training compared to controls having received the same training.
2nd model, as 1st only added the difference in task performance (demeaned, after-before) as covariate in a single column:
contrast: patient [1 0] shows some differences in regions comparable to 1st model, though not as strong
contrast control [0 1] is empty
contrast: patients - control [1 -1 0] now shows no differences
contrast for difference in task performance [0 0 1] is blank as well
I am not sure what to learn from this, probably simply that after adjusting for task performance (maybe taking out the training effect in the actual scanning task which was a different one to the training-interestingly only in the patient group, so maybe something else)
3rd model, as 2nd only splitting the covariates (demeaned) into the respective (2) columns to test for interactions
contrast: patient>control [0 0 1 -1] is blank
contrast: patient<control [0 0 -1 1] shows some differences in regions with little overlap but more adjacent to patient contrast [1 0] from 2nd model
contrasts [1 0 0 0] and [1 -1 0 0] show some differences though, but I understand that they are not interpretable.
My laymans interpretation of the interaction in [0 0 -1 1] is that brain activation is stronger for controls (for the difference after-before) than for patients as differences in task performance (after-before) increase
One question:
As mentioned above, I understand that main effects shouldn't be interpreted when interactions are significant (model 3). But is it legitimate to interpret the results from the 1st and 2nd model knowing that there are interactions (3rd model).
Any comments on the above are very welcome!
Thanks again for pointing me into the right direction here!
All the best,
Torsten
