Thank you Donald.  That was my thinking. 

The other issue is with regards to the fixed/mixed effects. Running the difference images using the fixed effects makes perfects sense. The question becomes a little more confusing at the next level. We only have good data from 5 subjects. I am not sure that it makes sense to generalize beyond these subjects. And, given this small subject number, detecting effects will be more difficult using a mixed effects model at this level. On one hand, I am sensitive to the likelihood of committing a type II error if I use the mixed effects. But, on the other hand, I need to balance this with committing a type I error using the fixed effects. 
Any thoughts on this?

Of course, it would be wonderful if we had data from many more subjects. Unfortunately, that is not possible. It would be a shame to waste the information that is in this data set.

Regards,

Rinah

On Wed, 21 Mar 2012 13:56:32 -0400, MCLAREN, Donald <[log in to unmask]> wrote:

>In this simple case, I would create the difference images of drug-placebo
>and then do a one-sample t-test with the covariate. The covariate effect
>will show where the drug level influences the change in activity.
>
>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
>Website: http://www.martinos.org/~mclaren
>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 Wed, Mar 21, 2012 at 10:44 AM, Rinah Yamamoto <
>[log in to unmask]> wrote:
>
>> Hi FSL experts,
>>
>> I have a question regarding an event related phMRI with covariate. We have
>> a group of subjects that received IV drug on one occasion and placebo on
>> another. We created individual regressors using their individual blood
>> levels of drug (interpolated curve across all volumes and z transformed).
>>
>> We have run the higher level(s) a number of ways (with arguments ensuing
>> regarding the most appropriate and interpretable method). One issue of note
>> is that this is a very small number of subjects (5).
>> A number of things are of interest. The simple dependent t (activation of
>> drug greater than placebo). The dependent t with a covariate. Here we have
>> a choice of two highly correlated covariates - one is the peak drug value
>> (demeaned?), the other is a measure of drug exposure history (demeaned?).
>>
>> One method we have tried is a two level higher level analysis. First using
>> lower level feat directories to get the paired copes for input to the next
>> level.
>> group    EV1  s1     s2    etc             with contrast for EV1 = +1 all
>> other EVs =0
>>     1          1       1.0     0                    using a fixed effects
>> model
>>     1        -1.0      1.0     0
>>     1         1        0       1.0
>>     1      -1.0        0       1.0
>>
>> These copes are then used at the next level with group = 1 for all and EV1
>> = 1 for all.  This will tell us how much these subjects activated to the
>> drug on average (having already accounted for the difference between drug
>> and placebo at the previous level).
>> This can then be run by itself or with one of the covariates - demeaned or
>> not???
>> One question is whether this level should be a mixed (flame 1) or fixed
>> design, given that there are only 5 subjects AND it may not make sense to
>> generalize beyond this group of subjects.
>>
>> ALTERNATIVELY
>> We have run these subjects with just one higher level
>>
>> group    EV1   s1     s2    etc             with contrast for EV1 = +1
>> (and -1 if desired) all other EVs =0
>>     1         1        1.0     0                    using a mixed effects
>> model (again question re: fixed or not given above mentioned issues)
>>     1         1        0       1.0
>>     1        -1.0      1.0     0
>>     1        -1.0      0       1.0
>>
>> This will tell us how much activation for drug>placebo (or vice versa)
>> However, adding in the covariate is a little more confusing. We don't have
>> a peak drug measure during the placebo scan (subjects have a baseline only
>> and were abstinent before the scan), so this can be 0. With both active and
>> placebo in the same design, do we demean the covariate as one group, ie
>> include the zeros, or separate the runs, demeaning just the active and
>> zeros for the placebos? If we use the exposure history as a covariate, the
>> measure is the same for both runs.
>>
>> Is there some way to make sense of all of this?
>> It is interesting to see the activation in response to the drug. But, it
>> is also interesting to note that the peak blood measures of drug were very
>> different in these subjects (given the same dose/body weight). And that
>> this was highly correlated with their exposure history. This is a drug that
>> definitely crosses the BBB very rapidly. It might be expected that
>> subjects' activation would be affected by their individual exposure/blood
>> levels.
>>
>> Thank you so much.
>>
>> Regards,
>>
>> Rinah
>>
>