Covariates are automatically mean-centered by SPM. The proof is in the
pudding so go to the directory in which you have a SPM.mat file and type
the following:
load SPM
mean(SPM.xX.X)
Each column refers to a column in your design matrix. You will notice
that any covariates you have entered have a mean of 0.
Regards,
Jejo
Nici Wenderoth wrote:
> Hi Jeff,
>
> this is more a question for the statistic experts (that will hopefully
> come
> in if I'm writing nonsense), but here my basic understanding of these
> thing:
>
> If you enter movement parameters to the model, you only want to remove
> movement related variance from your data. For that purpose you don't have
> to care about mean corrections, because you don't want to do any further
> statistics with it.
>
> This is different if you are interested in e.g. the correlation of brain
> activation and another variable such as perception scores. Let's say you
> have 2 conditions (A and B), each performed for 3 perception levels with
> the perception scores [15 10 5] for A and [36 30 24] for B. If you enter
> the scores as they are, you are looking for areas that
> 1. show a linear decrease over the 3 perception levels (as indicated
> by the
> scores) and
> 2. are more strongly activated in B than in A (since the mean scores are
> higher for B than for A).
> Thus, you are mixing up perception and condition effects, what you
> probably
> don't want to do. You don't have this problem, if you enter mean
> corrected
> perception scores and model the condition effect of A and B separately.
>
> I'm not so sure whether this example is really good, but it maybe
> gives you
> an idea?
>
> Regards, Nici
>
>
>
>
>> Dear spm-maillist and Nici
>>
>> Do movement confounds then have to be centered around 0 or their mean??
>>
>> Unless I am thinking wrong, the realigned image volumes are commonly
>> coregistered to the first image volume rather than the mean image
>> volume.
>> The former creates confounds that do not center around 0 whereas the
>> latter do if I am thinking correctly. Are the former a problem for
>> entering as use specificied regression in the fMRI design step??
>>
>> Even the famous faces / nonfamous faces example on
>> http://www.fil.ion.ucl.ac.uk/~wpenny/datasets/face-rep/SPM2/README-SPM2.txt
>>
>>
>> coregisters to the first image volume rather than the mean one.
>>
>> Aren't hrfs often not mean centerted 0 as most of the activity is above
>> baseline?
>>
>> Thanks
>> Jeff Lorberbaum
>>
>>
>>
>> On Sat, 13 Dec 2003, Nici Wenderoth wrote:
>>
>> > At 20:28 12.12.03 +0000, you wrote:
>> >
>> > Hi Kyle
>> >
>> > >Dear SPM list,
>> > >
>> > > I conducted an event-related study (TR=2.0 sec) in which
>> > >participants looked at pictures of common objects during three
>> scanning
>> > >runs. I later had a separate group of participants rate the
>> stimuli on
>> > >several dimensions (e.g., how easily a picture can be imagined).
>> I now
>> > >want to see if the imagery ratings predict the intensity of activity
>> within
>> > >the clusters of activity in the standard fixed effects analysis of
>> the
>> FMRI
>> > >data. I assume that this information should be entered as a
>> user-specified
>> > >regressor in the design matrix. I have two questions though:
>> > >
>> > >First, should I mean correct the vector describing the imagery
>> ratings?
>> > >The problem with mean correcting the scores is that some of the
>> values are
>> > >now negative. In this case, I expect that activity should
>> increase with
>> > >each picture presentation, but that it should be greater based on how
>> > >easily it can be imagined. When I mean correct, half of the
>> scores are
>> > >negative, which might suggest deactivation for a given stimulus
>> > >presentation. While researchers often mean correct their
>> regressors, is
>> > >mean correction always necessary?
>> >
>> >
>> > Yes, you should mean correct your regressors (as repeatedly
>> suggested by
>> > the spm list). In this analysis you are looking at the correlation
>> between
>> > your perception score and brain activity. That means for a positive
>> > correlation that brain activation should be low when the perception
>> scores
>> > are low and high when the perception scores are high. Thus, for the
>> > correlation it does not matter whether scores are partly negative
>> or not.
>> >
>> >
>> > >Second, if I do mean correct the scores, should I use the mean
>> rating of
>> > >the stimuli within the run, or should I use the overall mean rating
>> for the
>> > >stimuli? In other words, should I use the same mean value in all
>> three
>> > >runs, or should I use a calculate a mean rating for just the
>> stimuli in a
>> > >given run, and subtract that value from each individual picture's
>> rating in
>> > >that particular run?
>> >
>> > mhhhhh, if you expect run specific effects (in other words, if you
>> > calculate an ANOVA on the perception scores, is there a significant
>> effect
>> > of run?), I would enter the mean rating of the stimuli within the
>> run. If
>> > not, I would average the values across runs to reduce the influence of
>> > outliers. however, i guess there could be more arguments in favour
>> for the
>> > one or the other procedure.
>> >
>> > Regards, Nici
>> >
>> >
>> > >Thanks for any suggestions!
>> >
>> > Dr. Nici Wenderoth
>> > Motor Control Laboratory
>> > Dep. of Kinesiology
>> > K.U.Leuven
>> > Tervuurse Vest 101
>> > 3001 Heverlee
>> > Belgium
>> >
>> > E mail: [log in to unmask]
>> > Tel: 32 16 32 90 72
>> > Fax: 32 16 32 91 97
>> >
>
>
> Dr. Nici Wenderoth
> Motor Control Laboratory
> Dep. of Kinesiology
> K.U.Leuven
> Tervuurse Vest 101
> 3001 Heverlee
> Belgium
>
> E mail: [log in to unmask]
> Tel: 32 16 32 90 72
> Fax: 32 16 32 91 97
|